Plant-microbial interactions in agriculture and the use of farming systems to improve diversity and productivity

Suzanne L. Ishaq; Suzanne L. Ishaq

doi:10.3934/microbiol.2017.2.335

AIMS Microbiology

2017, Volume 3, Issue 2: 335-353. doi: 10.3934/microbiol.2017.2.335

Previous Article Next Article

Review Special Issues

Plant-microbial interactions in agriculture and the use of farming systems to improve diversity and productivity

Suzanne L. Ishaq ^,

Department of Land Resources and Environmental Sciences, Montana State University, Bozeman, MT 59717, Montana, USA

Received: 15 March 2017 Accepted: 04 May 2017 Published: 11 May 2017

A thorough understanding of the services provided by microorganisms to the agricultural ecosystem is integral to understanding how management systems can improve or deteriorate soil health and production over the long term. Yet it is hampered by the difficulty in measuring the intersection of plant, microbe, and environment, in no small part because of the situational specificity to some plant-microbial interactions, related to soil moisture, nutrient content, climate, and local diversity. Despite this, perspective on soil microbiota in agricultural settings can inform management practices to improve the sustainability of agricultural production.

Keywords:

Citation: Suzanne L. Ishaq. Plant-microbial interactions in agriculture and the use of farming systems to improve diversity and productivity[J]. AIMS Microbiology, 2017, 3(2): 335-353. doi: 10.3934/microbiol.2017.2.335

Related Papers:

[1]	Hongli Niu, Kunliang Xu . A hybrid model combining variational mode decomposition and an attention-GRU network for stock price index forecasting. Mathematical Biosciences and Engineering, 2020, 17(6): 7151-7166. doi: 10.3934/mbe.2020367
[2]	Hongli Niu, Yazhi Zhao . Crude oil prices and volatility prediction by a hybrid model based on kernel extreme learning machine. Mathematical Biosciences and Engineering, 2021, 18(6): 8096-8122. doi: 10.3934/mbe.2021402
[3]	Biyun Hong, Yang Zhang . Research on the influence of attention and emotion of tea drinkers based on artificial neural network. Mathematical Biosciences and Engineering, 2021, 18(4): 3423-3434. doi: 10.3934/mbe.2021171
[4]	Hongji Xu, Shi Li, Shidi Fan, Min Chen . A new inconsistent context fusion algorithm based on BP neural network and modified DST. Mathematical Biosciences and Engineering, 2021, 18(2): 968-982. doi: 10.3934/mbe.2021051
[5]	Eunjae Choi, Yoosang Park, Jongsun Choi, Jaeyoung Choi, Libor Mesicek . Forecasting of garlic price based on DA-RNN using attention weight of temporal fusion transformers. Mathematical Biosciences and Engineering, 2023, 20(5): 9041-9061. doi: 10.3934/mbe.2023397
[6]	Xin Jing, Jungang Luo, Shangyao Zhang, Na Wei . Runoff forecasting model based on variational mode decomposition and artificial neural networks. Mathematical Biosciences and Engineering, 2022, 19(2): 1633-1648. doi: 10.3934/mbe.2022076
[7]	Xiangyang Ren, Juan Tan, Qingmin Qiao, Lifeng Wu, Liyuan Ren, Lu Meng . Demand forecast and influential factors of cold chain logistics based on a grey model. Mathematical Biosciences and Engineering, 2022, 19(8): 7669-7686. doi: 10.3934/mbe.2022360
[8]	Xiong Wang, Zhijun Yang, Hongwei Ding, Zheng Guan . Analysis and prediction of UAV-assisted mobile edge computing systems. Mathematical Biosciences and Engineering, 2023, 20(12): 21267-21291. doi: 10.3934/mbe.2023941
[9]	Xiaotong Ji, Dan Liu, Ping Xiong . Multi-model fusion short-term power load forecasting based on improved WOA optimization. Mathematical Biosciences and Engineering, 2022, 19(12): 13399-13420. doi: 10.3934/mbe.2022627
[10]	Hongqiang Zhu . A graph neural network-enhanced knowledge graph framework for intelligent analysis of policing cases. Mathematical Biosciences and Engineering, 2023, 20(7): 11585-11604. doi: 10.3934/mbe.2023514

Abstract

1. Introduction

The stock price is the primary factor affecting investors. Short-term stock price forecasts have important significance for reference to investors. The factors affecting stock price are very complex, which makes it difficult to analyze and predict stock price. The stock market can be seen as a nonlinear dynamic system under the action of multiple factors. Time series forecasting models are usually adopted for stock price forecasts. There are multiple regression models, exponential smoothing models, ARIMA models, neural network models, combined forecasting models, and grey forecasting models.

Many scholars have conducted a series of studies on stock price prediction. Shi ^[1] used ARMA and BP neural networks to obtain the linear and non-linear components of the stock price sequence and then used the Markov model to correct the error of the result. Vijh ^[2] used the random forest algorithm to integrate the indicators of multiple stocks, and then used BP neural network to output the predicted observations. Liu ^[3] proposes a new effective metric, Mean Profit Rate (MPR). The effectiveness of metrics is measured based on the correlation between the metric value and profit of the model. And, the relationship between the features of Bitcoin and the next day change in the price of Bitcoin using an Artificial Neural Network ensemble approach called Genetic Algorithm based Selective Neural Network Ensemble ^[4] is explored by Sin. Zhu ^[5] proposes an intelligent trading system using support vector regression optimized by genetic algorithms (SVR-GA) and multilayer perceptron optimized with GA (MLP-GA). And, an improved one-step ahead forecasting system and an improved multi-step ahead prediction ^[6] are proposed by Dong. Zhang ^[7] uses BP neural network to train stock prices in the form of spline division, and then output the predicted observations with data independent of the training samples. Teo ^[8] train the wavelet packet multi-layer perceptron neural network (WP-MLP) by backpropagation for time series prediction. Summarizing the above literature, using deep learning and machine learning methods to predict stock prices can achieve good prediction results. However, most methods do not consider the memory of the sequence when processing the original price sequence. Sang ^[9] studied the characteristics of long-term memory of fluctuations in the Chinese stock market, which means that the historical data of the stock market have an impact on the stock price at the current time. Since fractional calculus has power-law memory ^[10], the idea of fractional calculus can be used to obtain information about historical data. Based on the characteristics of fractional calculus, some models that combined the idea of fractional calculus with traditional time series forecasting methods have attracted the attention of many scholars. For example, the ARFIMA model, fractional grey model, etc. are the models that combine fractional ideas. Because the grey model has good performance in the uncertainty system, the prediction of the stock price is very consistent with the grey model.

The Grey model is an important part of the grey theory, and the most famous one is GM (1, 1) model. However, GM (1, 1) has many shortcomings. The background value is based on approximation, and it cannot handle high-dimensional data. And the ability of nonlinear fitting is insufficient. Because of the shortcomings, many scholars had improved the GM (1, 1) model ^{[11,12,13,14,15,16,17,18,19,20,21,22,23,24,25]}. Tien ^[13] proposed the GMC (1, m) model to solve the shortcomings of traditional GM (1, 1) that cannot handle high-dimensional data. Literature ^{[14,15,16,17,18,19]} proposed a variety of grey Bernoulli prediction models and grey models combined with kernel methods, which solved the shortcomings of insufficient nonlinear fitting capabilities. Literature ^[20,21] had solved the shortcoming that the background value is based on approximation. Literature ^{[22,23,24,25]} combined fractional calculus with the grey model and used historical data to solve many practical application problems.

Since most time series including stock price series usually contain linear and non-linear factors, a single forecasting model cannot achieve good results. According to the high-dimensional, memorable, and non-linear characteristics of stock data, a Hausdorff fractional grey model with convolution, locally linear embedding, and BP neural network (HFGMC (1, m)-LLE-BP) model is proposed in this paper. HFGMC (1, m) model is improved from FGMC (1, m) in literature ^[25]. The FGMC (1, m) model has several disadvantages. There are factorial operators in the elements of the accumulation generation matrix in the FGMC (1, m) model. When the data length is long, it will have the disadvantages of complex calculation and information loss. The calculation of background value in the FGMC (1, m) model is approximately obtained by using the trapezoidal formula between the two values in the sequence, which will calculate background value ignore the local change trend of the sequence. These shortcomings degrade the prediction performance of FGMC (1, m). The operator derived from HFGMC (1, m) model combined with Hausdorff fractional derivative does not contain a factorial operator, which solves the shortcomings of complex calculation and information loss. Because of the disadvantages of the background value based on approximation, the HFGMC (1, m) model uses the Newton-Cotes discrete formula and the Newton mean difference interpolation polynomial to optimize. Because the Newton-Cotes discrete formula and the Newton mean difference interpolation polynomial consider the changing trend of local data, it can solve the shortcomings of approximate background values. HFGMC (1, m) is used to predict the linear component of stock price. The BP neural network is used to predict the nonlinear component of stock price. The prediction result is obtained by adding a linear component and a nonlinear component.

This paper uses the stock closing price as the forecast target to verify the performance of HGFMC (1, m)-LLE-BP. Compared with FGMC (1, m), the prediction accuracy of HFGMC (1, m) is much higher than that of FGMC (1, m). This is because FGMC (1, m) only retains local information, while HGFMC (1, m) retains global information. Therefore, the prediction accuracy of HGFMC (1, m) will be higher. Besides, thanks to the excellent nonlinear fitting ability of the BP neural network, it can fit the nonlinear residuals in the time series. The prediction results show that the BP neural network improves the performance of HGFMC (1, m). Finally, compare the HGFMC (1, m)-LLE-BP model with the FGMC (1, m)-LLE-BP model, BP neural network model, LSTM model and the ARIMA-BP hybrid model. The performance of the HGFMC (1, m)- LLE-BP model is far better than other models.

The rest of this paper is organized as follows. In Section 2, the specific framework of the proposed model is given. In Section 3, the FGMC (1, m) is described. In Section 4, the HFGMC (1, m) is given. In Section 5, the LLE algorithm and BP neural network are introduced. In Section 6, experiments have verified the effectiveness of the proposed model. In Section 7, the work of this paper is summarized.

2. Specific frameworks of the proposed model

In this paper, the HFGMC (1, m) is proposed to predict the linear component of the stock price, and then the BP neural network is used to predict the nonlinear component of the stock price. Because stock price is affected by nonlinear high-dimensional stock data, this paper used the LLE algorithm to reduce the dimensionality of nonlinear high-dimensional stock data. The flow chart of the model proposed in this paper is shown in Figure 1.

Figure 1. The flow chart of the model proposed in this paper.

DownLoad: Full-Size Img PowerPoint

The text description of the specific steps is as follows.

Step 1: Input parameters r, K. The parameter r is used to calculate the r-order accumulative generator matrix. The parameter K represents the number of neighboring points of each sample point, which is used for the calculation of locally linear embedding.

Step 2: Input the original data matrix. Suppose the original data matrix ${X^{\left( 0 \right)}}'$ , $X' \in {R^{n \times m}}$ . n is the length of the sequence, and m is the m stock indicators including the sequence to be predicted. Select the sequence to be predicted (usually the first column of the original data matrix), and the remaining sequences are used as related factors that affect the sequence to be predicted (usually other columns in the original data matrix except for the first column, and each column represents a related factor).

Step 3: Form the correlation factor matrix that affects the sequence to be predicted, and apply the locally linear embedding algorithm to reduce the dimension to d dimension. The sequence to be predicted and the correlation factor matrix after dimensionality reduction form the input matrix ${X^{\left( 0 \right)}}$ . Then normalize the input matrix.

Step 4: Perform Hausdorff fractional accumulative generation operation (HFAGO) on the input matrix to obtain the Hausdorff fractional accumulative matrix.

Step 5: Combine the Newton-Cotes formula and use the Hausdorff fractional accumulation matrix to calculate the background value.

Step 6: After obtaining the optimized background value, combine the least square estimation to solve the parameters.

Step 7: Solve the whitening equation and get the time response function. Solve the in-sample time response sequence.

Step 8: Perform the inverse Hausdorff fractional accumulation generation operation (I-HFAGO) on the time response sequence within the sample to obtain the in-sample prediction sequence.

Step 9: Subtract the in-sample prediction sequence and the in-sample true value sequence to obtain the in-sample residual sequence. The mean of the sum of the residual sequence elements is the mean absolute error (MAE). Update r and K, and then judge whether the parameters r and K exceed the search range. If it exceeds the search range, select r, K that minimize MAE to solve the in-sample prediction sequence, and then go to step 10. If it does not exceed the search range, repeat Step 2 to Step 9.

Step 10: Input the out-of-sample correlation factor matrix R, and then use the time response function to solve the out-of-sample time response sequence. Then the out-of-sample time response function obtains the out-of-sample prediction sequence through I-HFAGO as a linear component.

Step 11: The in-sample residual sequence uses BP neural network to do nonlinear fitting, and then the out-of-sample residual sequence is output as the nonlinear component.

Step 12: Add the linear component and the non-linear component to get the combined prediction sequence.

The above steps can be divided into 6 modules. Step 1 to Step 3 is the data preprocessing part. Step 4 to Step 7 is the modeling process of the HFGMC (1, m). Step 8 to Step 9 is the residual calculation part. Step 10 is to calculate the linear component through the HFGMC (1, m). Step 11 is to calculate the non-linear part through BP neural network. Step 12 gives the prediction result.

3. The FGMC (1, m) model

Literature ^[25] gave the specific derivation process of FGMC (1, m). Let the original data matrix be ${X^{\left( 0 \right)}}$ , $X \in {R^{n \times m}}$ . n is the length of the sequence, and m is the indicators including the sequence to be predicted. Then perform the r-order accumulative generation operation (r-AGO) on each column of the original data matrix to obtain the r-order accumulative generation sequence. The mathematical form is defined as

$X_j^{(r)} = \{ x_j^{(r)}(1), x_j^{(r)}(2), \ldots , x_j^{(r)}(n)\} , j = 1, 2, \ldots , m$

(1)

$x_j^{\left( r \right)}(t) = \sum\limits_{i = 1}^t {x_j^{\left( 0 \right)}(i)\left( {\begin{array}{*{20}{c}} {t - i + r - 1} \\ {t - i} \end{array}} \right)}$

(2)

where $\left( {\begin{array}{*{20}{c}} {t - i + r - 1} \\ {t - i} \end{array}} \right){\rm{ = }}\frac{{\left( {t - i + r - 1} \right)\left( {t - i + r - 1} \right)...\left( {r + 1} \right)r}}{{\left( {t - i} \right)!}}{\rm{ }}$ , and r is called the fractional order. r should be taken in the interval (0, 1). However, r can be an arbitrary real number. The matrix form of the above equations can be written as

$X_j^{\left( r \right)} = {A^r}_{n \times n}X_j^{\left( 0 \right)}$

(3)

where ${A^r}_{n \times n}$ is called the r-order accumulative matrix, and its form is

$A_{n \times n}^r = {\left[ {\begin{array}{*{20}{c}} 1&{}&{}&{}&{} \\ r&1&{}&{}&{} \\ {\frac{{r\left( {r + 1} \right)}}{{2!}}}&r&1&{}&{} \\ \vdots & \vdots & \vdots & \ddots &{} \\ {\frac{{r\left( {r + 1} \right) \cdots \left( {r + n - 2} \right)}}{{\left( {n - 1} \right)!}}}&{\frac{{r\left( {r + 1} \right) \cdots \left( {r + n - 3} \right)}}{{\left( {n - 2} \right)!}}}& \cdots &r&1 \end{array}} \right]_{n \times n}}$

(4)

Then establish the whitening equation. The specific mathematical form is

$\frac{{dx_1^{\left( r \right)}\left( {\tau + t} \right)}}{{dt}} + {b_1}x_1^{\left( {\rm{r}} \right)}\left( {\tau + t} \right) = \sum\limits_{j = 2}^m {b{}_jx_j^{\left( r \right)}\left( t \right)} + u$

(5)

where $\tau$ is the delay period and u is the control coefficient. When $\tau$ and u are equal to 0, the traditional GM (1, m) model can be obtained. But the GM (1, m) had been proved inaccurate in ^[13]. When $\tau$ is equal to 0 and m is equal to 1, it can degenerate into a GM (1, 1) model.

By integrating the Eq (5) and using the trapezoidal formula in the interval [k-1, k], the discrete form of the existing fractional grey model can be obtained as

$\left( {x_1^{\left( r \right)}\left( {\tau + t} \right) - x_1^{\left( r \right)}\left( {\tau + t - 1} \right)} \right) + {b_1}Z_1^{\left( r \right)}\left( {\tau + t} \right){\rm{ = }}\sum\limits_{j = 2}^m {{b_j}Z_j^{\left( r \right)}(t)} + u$

(6)

where $Z_j^{\left( r \right)}$ is called the background values, which are defined as

$Z_1^{\left( r \right)}\left( {\tau + t} \right) = \frac{1}{2}\left[ {X_1^{\left( r \right)}\left( {\tau + t} \right) + X_1^{\left( r \right)}\left( {\tau + t - 1} \right)} \right]$

(7)

${Z}_{j}^{\left(r\right)}\left(t\right) = \frac{1}{2}\left[{X}_{j}^{\left(r\right)}\left(t\right)+{X}_{j}^{\left(r\right)}\left(t-1\right)\right],j = 2, 3, \dots , m$

(8)

The parameter ${b_1}, {b_2}, \ldots , {b_m}, u$ of Eq (5) can be obtained by least square estimation. Then substitute it into the whitening equation of Eq (4), and solve the differential equation to obtain the time response function. The form of the time response function is

$\hat x_{\rm{1}}^{\left( r \right)}\left( {\tau + t} \right) = x_1^{\left( r \right)}\left( {\tau + 1} \right){e^{ - {b_1}\left( {t - 1} \right)}} + \int_1^t {{e^{ - {b_1}\left[ {t - 1 - \left( {\upsilon - 1} \right)} \right]}}\left[ {\sum\limits_{j = 2}^m {{b_j}x_j^{\left( r \right)}\left( \upsilon \right)} + u} \right]} d\upsilon$

(9)

The convolution operation of the second term on the right side of the Eq (9) can be discretized using the Gaussian formula to obtain the following discrete equation. The specific mathematical form is

$\hat x_{\rm{1}}^{\left( r \right)}\left( {\tau + t} \right) = x_1^{\left( r \right)}\left( {\tau + 1} \right){e^{ - {b_1}\left( {t - 1} \right)}}{\rm{ + }}\sum\limits_{\upsilon = 2}^t {\left\{ {{e^{ - {b_1}\left( {t - \upsilon + \frac{1}{2}} \right)}} \cdot \frac{1}{2}\left[ {f\left( \upsilon \right) + f\left( {\upsilon - 1} \right)} \right]} \right\}}$

(10)

where $f\left( \upsilon \right) = \sum\limits_{j = 2}^m {{b_j}x_j^{\left( r \right)}\left( \upsilon \right)} + u$ . Finally, perform the inverse r-order accumulative generation operation (r-IAGO) to obtain the predicted sequence

$\hat X_{\rm{1}}^{\left( {\rm{0}} \right)} = {\left( {{A^r}} \right)^{ - 1}}\hat X_1^{\left( r \right)}$

(11)

4. The HFGMC (1, m)

Accumulate generation operation (AGO) is the core step of the grey model, which can eliminate the instability caused by fluctuations in the original data. From the r-order accumulation matrix in Eq (4), it can be seen that the original sequence needs to be multiplied by a factorial multiplier when the accumulation operation is performed. When the length of the sequence is long, the calculation in the r-order accumulation matrix will be very complicated. Besides, it can be seen from Eqs (7) and (8) that the calculation of the background value is only obtained by using a simple trapezoidal calculation formula. However, most systems do not grow linearly, which leads to calculation errors when using FGMC (1, m) models. The error is shown in Figure 2.

Figure 2. The flow chart of the model proposed in this paper.

DownLoad: Full-Size Img PowerPoint

4.1. Derivation of Hausdorff Fractional Accumulate Generation Operation (HFAGO)

The Hausdorff fractional derivative is defined as ^[26].

$\frac{{df\left( x \right)}}{{d{x^r}}} = \mathop {\lim }\limits_{{x'} \to x} \frac{{f\left( x \right) - f\left( {{x'}} \right)}}{{{x^r} - {x^{'r}}}}$

(12)

Since the first difference can usually be approximately defined by the first derivative as

$\Delta f\left( t \right) \approx \mathop {\lim }\limits_{h \to 1} \frac{{f\left( x \right) - f\left( {x - h} \right)}}{h}{|_{x = t}} = f\left( t \right) - f\left( {t - 1} \right)$

(13)

Therefore, the difference of Hausdorff fractional derivative can be approximately defined by Hausdorff fractional derivative as

${\Delta ^r}f\left( t \right) \approx \mathop {\lim }\limits_{{x'} \to \left( {x - 1} \right)} \frac{{f\left( x \right) - f\left( {x'} \right)}}{{{x^r} - x{'^r}}}{|_{x = t}} = \frac{{f\left( t \right) - f\left( {t - 1} \right)}}{{{t^r} - {{\left( {t - 1} \right)}^r}}} = {[{t^r} - {\left( {t - 1} \right)^r}]^{ - 1}}\Delta f\left( t \right)$

(14)

Let $f\left( t \right) = {x^{\left( 0 \right)}}\left( 1 \right), f\left( {t - 1} \right) = {x^{\left( 0 \right)}}\left( 2 \right), \ldots , f\left( 1 \right) = {x^{\left( 0 \right)}}\left( t \right)$ (This is the reverse mapping of the sequence when doing AGO operations), then 1-order reverse accumulate operation (1-RAGO) can be defined as

$\nabla f\left( t \right) = \sum\limits_{j = 1}^t {f\left( j \right)}$

(15)

1-RAGO is essentially the same as 1-AGO because it just does a simple inverse mapping of the sequence.

$\nabla f\left( t \right) = \sum\limits_{j = 1}^t {f\left( j \right) = {x^{\left( 1 \right)}}\left( t \right)} = \sum\limits_{i = 1}^t {{x^{\left( 0 \right)}}} \left( i \right)$

Eq (15) combined with Eq (13) to get

$\Delta \nabla f\left( t \right) = \Delta \left( {\sum\limits_{j = 1}^t {f\left( j \right)} } \right) = \sum\limits_{j = 1}^t {f\left( j \right)} - \sum\limits_{j = 1}^{t{\rm{ - }}1} {f\left( j \right)} = f\left( t \right)$

(16)

In the background of the grey prediction model, the difference and accumulate operation are reversible ^[24]. Eq (16) can be also written as

$\nabla \Delta f\left( t \right) = f\left( t \right)$

(17)

Without loss of generality, HFAGO is expressed as ${\nabla ^r}$ and satisfies the following relationship

${\Delta ^r}{\nabla ^r}f\left( t \right) = f\left( t \right)$

(18)

Combine Eq (17) with Eq (14), then move the term to the right to get

$\Delta \left( {{\nabla ^r}f\left( t \right)} \right) = f\left( t \right) \cdot [{t^r} - {\left( {t - 1} \right)^r}]$

(19)

Combining Eq (17), Eq (15) is simplified to

${\nabla ^r}f\left( t \right) = \sum\limits_{j = 1}^t {f\left( j \right)\left[ {{j^r} - {{\left( {j - 1} \right)}^r}} \right]}$

(20)

Equation (20) is HFAGO. Compared with Eq (2), HFAGO does not contain a factorial multiplier. To understand the difference between r-AGO and HFAGO more intuitively, the next step is to visually compare r-AGO and HFAGO. Assuming

$\phi \left( t \right) = \frac{{\Gamma \left( {r + t - 1} \right)}}{{\Gamma \left( t \right) \cdot \Gamma \left( r \right)}}$

(21)

$\omega \left( t \right) = {t^r} - {\left( {t - 1} \right)^r}$

(22)

where $t$ is equal to $1, 2, \ldots , n$ . The $\Gamma \left( \bullet \right)$ in Eq (21) is the gamma function. Using Eqs (21) and (22), Eq (4) can be rewritten as

${A^r} = {\left[ {\begin{array}{*{20}{c}} {\phi \left( 1 \right)}&{}&{}&{}&{} \\ {\phi \left( 2 \right)}&{\phi \left( 1 \right)}&{}&{}&{} \\ {\phi \left( 3 \right)}&{\phi \left( 2 \right)}&{}&{}&{} \\ \vdots & \vdots & \ddots &{\phi \left( 1 \right)}&{} \\ {\phi \left( n \right)}&{\phi \left( {n - 1} \right)}& \cdots &{\phi \left( 2 \right)}&{\phi \left( 1 \right)} \end{array}} \right]_{n \times n}}$

(23)

$A_H^r = {\left[ {\begin{array}{*{20}{c}} {\omega \left( 1 \right)}&{}&{}&{}&{} \\ {\omega \left( 2 \right)}&{\omega \left( 1 \right)}&{}&{}&{} \\ {\omega \left( 3 \right)}&{\omega \left( 2 \right)}&{}&{}&{} \\ \vdots & \vdots & \ddots &{\omega \left( 1 \right)}&{} \\ {\omega \left( n \right)}&{\omega \left( {n - 1} \right)}& \cdots &{\omega \left( 2 \right)}&{\omega \left( 1 \right)} \end{array}} \right]_{n \times n}}$

(24)

Assuming that r = 0.5, visualized Eqs (21) and (22). The visual comparison is shown in Figure 3.

Figure 3. Visual comparison chart of Eqs (21) and (22).

DownLoad: Full-Size Img PowerPoint

In Figure 3, the solid line represents the change of the multiplier overtime when r-AGO is performed. The dotted line represents the change of the multiplier over time when performing HFAGO. The analysis shows that the r-AGO is 0 when t > 172. This is because the factorial number is too large, so calculate the value of the denominator in matlab2020a will get INF. It will result in the loss of subsequent information and make the calculation result inaccurate. The visualization curve of HFAGO is very similar to the traditional r-AGO. On this basis, HFAGO can retain global information and reduce the amount of calculation.

4.2. Optimization of background value based on Newton-Cotes formula

Integrate both sides of Eq (5) on the interval [k-1, k] at the same time.

$\left( {x_1^{\left( r \right)}\left( {\tau + t} \right) - x_1^{\left( r \right)}\left( {\tau + t - 1} \right)} \right) + {b_1}\int_{k - 1}^k {x_1^{\left( r \right)}} \left( {\tau + t} \right)dt = \sum\limits_{j = 2}^m {{b_j}} \int_{k - 1}^k {x_j^{\left( r \right)}\left( t \right)dt + u}$

(25)

To optimize the error caused by the trapezoidal formula, it is necessary to use the Newton-Cotes formula to calculate the optimized background value for the integral term in Eq (25). Assume the integration area [a, b] is divided into l equal parts. Select node x_i = a+ih to construct a discrete integral formula, where h is the step size. Then, the constructed discrete integral formula is called the Newton-Cotes formula. The mathematical form is defined as

${I_l} = \left( {b - a} \right)\sum\limits_{k = 0}^l {C_i^{\left( l \right)}f\left( x \right)}$

(26)

The C^(l) in Eq (26) is called the Cotes coefficient. When l = 1, it is a trapezoidal formula. when l = 4, the corresponding quadrature formula is

${I_4} = \frac{{b - a}}{{90}}\left[ {7f\left( a \right) + 32f\left( {a + \frac{1}{4}} \right) + 12f\left( {a + \frac{1}{2}} \right) + 32f\left( {a + \frac{3}{4}} \right) + 7f\left( b \right)} \right]$

(27)

This paper divided the integration area into 4 equal parts. Then, the integral term in Eq (25) is constructed using Eq (27) to obtain the optimized background value. The optimized background value is

${Z^{\left( r \right)}}\left( {k + 1} \right) = \frac{1}{{90}}\left[ {7{x^{\left( r \right)}}\left( k \right) + 32{x^{\left( r \right)}}\left( {k + \frac{1}{4}} \right) + 12{x^{\left( r \right)}}\left( {k + \frac{1}{2}} \right) + 32{x^{\left( r \right)}}\left( {k + \frac{3}{4}} \right) + 7{x^{\left( r \right)}}\left( {k + 1} \right)} \right]$

(28)

It can be noticed that ${x^{\left( r \right)}}\left( {k + \frac{1}{4}} \right)$ , ${x^{\left( r \right)}}\left( {k + \frac{1}{2}} \right)$ , ${x^{\left( r \right)}}\left( {k + \frac{3}{4}} \right)$ do not exist for discrete data, and the Newton means difference interpolation polynomial can be used to solve this problem. The form of the Newton means difference interpolation polynomial is

${N_n}\left( x \right) = f\left( {{x_0}} \right) + f\left[ {{x_0}, {x_1}} \right]\left( {x - {x_0}} \right) + \ldots + f\left[ {{x_0}, {x_1}, \ldots , {x_n}} \right]\left( {x - {x_0}} \right)\left( {x - {x_1}} \right)\left( {x - {x_n}} \right)$

(29)

where $f\left[ {{x_0}, {x_1}, \ldots , {x_n}} \right]$ is the (n-1) order mean difference of the function $f\left( x \right)$ , and n represents the data points involved in the construction of the interpolation polynomial. However, the more points involved in the construction of the polynomial are not the better. When n > 8, the Runge effect usually appears ^[27]. This paper will use every 4 data points to construct a Newton mean difference interpolation polynomial.

5. Local linear embedding algorithm and BP neural network

In this section, the LLE algorithm ^[28] and BP neural network ^[29] are introduced.

5.1. Local linear embedding (LLE) dimensionality reduction algorithm

The LLE algorithm was proposed by S.T. Roweis and L.K. Saul in 2000. The LLE algorithm is an important dimensionality reduction algorithm in flow learning. The principle is to assume that the data in a high-dimensional space is linear in a region, and certain data can be expressed linearly using data samples in its neighborhood. Then map the data to a low-dimensional space, and try to keep the weight coefficients of the linear relationship before and after the projection the same.

The algorithm steps are as follows.

Step 1: Find the K neighboring points of each sample point ${X_i}$ in the high-dimensional space, and then calculate the Euclidean distance between them.

Step 2: Solve the covariance matrix between neighboring points.

Step 3: Solve the weight coefficient vector of each sample point.

Step 4: The weight coefficient vector forms the weight coefficient matrix W, and then calculates the matrix M that can minimize the reconstruction error function.

Step 5: Solve d+1 smallest eigenvalues in matrix M. The eigenvector corresponding to the eigenvalues from 2 to d+1 are the original data matrix X corresponding to the mapping space with the smallest reconstruction error.

5.2. BP neural network

BP neural network was proposed by Hinton and McClelland in 1986. The core idea is based on the multi-layer forward neural network combined with the error loss gradient descent direction. The error loss gradient descent direction is combined with the error backpropagation algorithm to update the weights of the multilayer neural network. After enough iterations, the error loss converges to the minimum and finally can approximate any nonlinear function.

6. Experiment analysis

Through the above method and combined with the process in Figure 1, experiments are carried out to verify the performance of the model proposed in this paper. The first step is to select experimental examples. The second step is to select parameters. The third step is to compare the predicted value with the real value. The last step is to evaluate the performance of the model proposed in this paper through error indicators.

6.1. Experimental examples

This paper selected the stock closing price as the forecast target and forecasted the closing price for the next 10 days to verify the performance of the model proposed in this paper. This paper divides the predicted stocks into two categories. The first category is stocks with large fluctuations in the closing price of the stock. For example, within the range of a given trading day (excluding public holidays), The stocks with the highest closing price minus the lowest closing price greater than 5 are the first type of stocks. The second category is stocks with small fluctuations in the closing price of the stock. For example, within the range of a given trading day, the stocks with the highest closing price minus the lowest closing price less than 2 are the second type of stocks. In order to verify the effectiveness of this model for the above two types of stocks, this paper selects 6 stocks from the Shanghai Stock Exchange (code prefix is SH) and Shenzhen Stock Exchange (code prefix is SZ) as prediction objects. The 6 stocks are, SZ002607, SZ002567, SZ000796, SH600586, SZ002770 and SZ000573. SZ002607, SZ002567 and SZ000796 belong to the first category of stocks. SH600586, SZ002770 and SZ000573 belong to the second category of stocks. Stock data were crawled on NetEase Finance and Economics website using a web crawler program.

Figure 4 shows the stock closing price trend chart for a total of 256 trading days from 2019-1-16 to 2020-2-10. The closing price of SH600586, SZ002770 and SZ000573 has a small variation range, while the closing price of SZ002607, SZ002567 and SZ000796 has a large variation range.

Figure 4. Daily closing prices of 6 stocks.

DownLoad: Full-Size Img PowerPoint

Table 1 shows the data used for modeling in SH600586, and the modeling data of the remaining 5 stocks are shown in the Supplementary. In this paper, the highest price, lowest price, opening price, previous closing price, floor change, floor, exchange, trading volume, trading amount, total market value and circulating market value are selected as indicators that affect the closing price of stocks. Indicators such as the highest price, lowest price, opening price, previous closing, floor change, floor and the exchange can reflect the changing trend of the closing price of stocks in the recent period. Indicators such as trading volume, trading amount, total market value and circulating market value can reflect the financial status of the stock company corresponding to the stock. Imitating the method of literature ^[30], the above indicators are reduced by dimensionality reduction algorithm, and then input into the HGFMC (1, m) model.

Table 1. Modeling data of SH600586.

No.	Date	Close	Date		Top_price	Low_price	Opening_price	Pre_price	Floor_price
1	19/1/16	2.95	19/1/2		3.01	2.94	2.96	2.95	-0.01
						$\vdots$
256	20/2/10	2.52	20/1/17		2.87	2.83	2.86	2.86	-0.02
			20/1/20		2.85	2.82	2.83	2.84	0.01
			20/1/21		2.86	2.82	2.84	2.85	-0.01
						$\vdots$
			20/2/10		2.53	2.43	2.46	2.47	0.05

No.	Date	Close	Date	Floor	Exchange	Volume	Amount	Total value	Circulation value
1	19/1/16	2.95	19/1/2	-0.339	0.5625	8,122,601	24,196,597	4,287,408,174	4,245,758,958
						$\vdots$
256	20/2/10	2.52	20/1/17	-0.6993	0.3202	4,574,400	13,015,088	4,057,706,800	4,057,706,800
			20/1/20	0.3521	0.3255	4,650,641	13,176,695	4,071,994,500	4,071,994,500
			20/1/21	-0.3509	0.3737	5,339,971	15,129,108	4,057,706,800	4,057,706,800
						$\vdots$
			20/2/10	2.0243	0.8219	11,743,314	29,233,159	3,600,500,400	3,600,500,400

| Show Table

DownLoad: CSV

6.2. Discussion of hyperparameters

First, the delay period parameter $\tau$ is discussed. It can be noted that there are 10 trading days' indicator data after the serial number 256. The indicator data of these 10 trading days is used to predict the closing price of the next 10 days. Since the HGFMC (1, m) model is a time series forecasting model that processes high-dimensional data, the model needs a priori data of the next 10 days to predict the stock closing price in the next 10 days. Considering that investors are not sensitive to changes in stock data ^[31] (except for market crises), this makes investors have a certain delay in responding to stock data. In other words, the stock data at the moment will have a time delay to affect investor sentiment, and then investor sentiment will affect the stock price. It can be seen from Eq (10) that HGFMC (1, m) integrates the delay function and shows the relationship between the stock price and various indicators. Therefore, when used Eq (10) to solve the time response function, the delay period is set to 10. The setting of the delay period $\tau$ is equal to the amount of prior data required.

Then the dimension d of the low-dimensional mapping space of the LLE algorithm is discussed. Literature ^[30,32,33] shows that setting the dimension d of the low-dimensional space to 2, 3, and 4 can well express the feature space of the data. Therefore, this paper uses dimension 3 as the dimension of the low-dimensional mapping space of the LLE algorithm.

6.3. Discussion on the stability of stock prices

The stationary test of stock prices is discussed in this section. This paper uses the ADF test to test the stability of stock prices.

The ADF test can check whether there is a unit root in the sequence. If there is a unit root, the sequence is not stationary. First, ADF test is performed on the original price series.

As shown in Table 2, the original sequence of 6 stocks cannot reject the existence of unit roots at the significance levels of 1, 5, and 10%. Therefore, the original sequence of 6 stocks can be considered unstable. Since the stock price series is not stationary, it is necessary to stabilize the series before forecasting.

Table 2. ADF test of the original sequence.

Stock code	Test statistic	Significance level
Stock code	Test statistic	1%	5%	10%
SZ000573	-1.34	-3.45	-2.87	-2.57
SZ000796	-2.22	-3.45	-2.87	-2.57
SZ002567	-1.62	-3.45	-2.87	-2.57
SZ002607	-2.10	-3.45	-2.87	-2.57
SZ002770	-0.75	-3.45	-2.87	-2.57
SH600586	-1.08	-3.45	-2.87	-2.57

| Show Table

DownLoad: CSV

From Eqs (1) and (20), it can be known that after the original sequence uses HFAGO, the random walk of the price appears as the increment between data points on the accumulative generation sequence. Therefore, the original sequence can be considered stationary after using HFAGO.

Figure 5 shows the stabilization process of SZ002770's stock price. The left image of Figure 5 shows the original sequence. The right figure of Figure 5 shows the accumulative generation sequence. The stationary test of the accumulative generation sequence of 6 stocks also uses the ADF test for further verification.

Figure 5. Stabilization of original series.

DownLoad: Full-Size Img PowerPoint

It can be seen from Table 3 that SZ000573, SZ002567, SZ002607, SZ002770 and SH600586 reject the hypothesis at the significance levels of 1, 5, and 10%, while SZ00796 rejects the hypothesis at the significance levels of 5% and 10%. Therefore, it can be considered that the accumulative generation sequence obtained by the original sequence after using HFAGO is stationary.

Table 3. ADF test of the accumulative generation sequence.

Stock code	Test statistic	Significance level
Stock code	Test statistic	1%	5%	10%
SZ000573	-5.46	-3.45	-2.87	-2.57
SZ000796	-2.87	-3.45	-2.87	-2.57
SZ002567	-6.52	-3.45	-2.87	-2.57
SZ002607	-6.31	-3.45	-2.87	-2.57
SZ002770	-3.66	-3.45	-2.87	-2.57
SH600586	-4.93	-3.45	-2.87	-2.57

| Show Table

DownLoad: CSV

6.4. Select parameters

It can be seen from Figure 1 that the establishment of the model requires the selection of appropriate parameters r and K. This paper established the contour diagram between the parameters r, K and the mean absolute error (MAE), and then found the parameter that could minimize the MAE.

Figure 6 is a contour diagram between the parameters r, K and MAE. In the figures, the horizontal axis represents the order r, and the vertical axis represents the number of K neighboring points of each sample point. The bluer the color, the smaller the MAE. The yellower the color, the larger the MAE. This paper enumerated the order r in the range of 0 to 1 in steps of 0.1 and enumerated the number of neighboring points K of the sample points from 0 to 100 in steps of 5. Then, select the search range that minimizes the MAE. Finally, use the exhaustive method within the range to get the relative optimal parameters.

Figure 6. ontour diagram of parameters r, K and MAE.

DownLoad: Full-Size Img PowerPoint

Table 4 shows the optimal parameter range of each stock. The search step for parameter r is 0.01, and the search step for parameter K is 5. For all stocks, the optimal search range of parameter K is 45 to 100. For SZ000573, SZ002770 and SH60586, the search range of parameter r is 0.25 to 0.6. For SZ000796, the search range of parameter r is 0.1 to 0.45. For SZ002567 and SZ002607, the search range of parameter r is 0.35 to 0.7.

Table 4. Parameter range.

Stock code	r	K
SZ000573	$r \in \left[ {0.25, 0.6} \right]$	$K \in \left[ {45,100} \right]$
SZ000796	$r \in \left[ {0.1, 0.45} \right]$
SZ002567	$r \in \left[ {0.35, 0.7} \right]$
SZ002607	$r \in \left[ {0.35, 0.7} \right]$
SZ002770	$r \in \left[ {0.25, 0.6} \right]$
SH600586	$r \in \left[ {0.25, 0.6} \right]$

| Show Table

DownLoad: CSV

Table 5 shows the result of parameters. For SZ000573, when the order r is 0.30 and K is 60, the MAE is 0.6120. For SZ000796, when the order r is 0.14 and K is 50, the MAE is 1.2227. For SZ002567, when the order r is 0.35 and K is 95, the MAE is 2.4767. For SZ002607, when the order r is 0.35 and K is 65, the MAE is 1.1929. For SZ002770, when the order r is 0.44 and K is 65, the MAE is 0.8013. For SH600586, when the order r is 0.41 and K is 55, the MAE is 0.6336.

Table 5. Result of parameters.

Stock code	r	K	MAE
SZ000573	0.30	60	0.6120
SZ000796	0.14	50	1.2227
SZ002567	0.35	95	2.4767
SZ002607	0.35	65	1.1929
SZ002770	0.44	65	0.8013
SH600586	0.41	55	0.6336

| Show Table

DownLoad: CSV

6.5. Discussion between HFGMC (1, m) and FGMC (1, m)

To make the distribution of the data input to FGMC (1, m) the same as HGFMC (1, m), the parameters of the data input to FGMC (1, m) also need to be set according to Table 5. The comparison of HFGMC (1, m) and FGMC (1, m) uses the mean error between the in-sample predicted sequence and the real sequence as an indicator. Table 6 shows the MAE comparison between HFGMC (1, m) and FGMC (1, m).

Table 6. The MAE comparison.

Stock code	FGMC (1, m)	HFGMC (1, m)
SZ000573	16.3341	0.6120
SZ000796	10.2270	1.2227
SZ002567	11.1814	2.4767
SZ002607	8.4681	1.1929
SZ002770	1.9685	0.8013
SH600586	3.0270	0.6336

| Show Table

DownLoad: CSV

It can be seen from Table 6 that the improvement of FGMC (1, m) by HGFMC (1, m) is very obvious, and the average MAE is reduced by 7.3778.

6.6. Discussion of BP Neural Network

In this section, the improvement of the model by the BP neural network is discussed. It can be known from Eq (10) that the out-sample forecast data can be regarded as a linear combination of various indicators, so a single HGFMC (1, m) can only predict the linear part of the future data. Due to the strong nonlinear fitting ability of the BP neural network, it can fit the residual function in the in-sample prediction sequence. Therefore, using BP neural network can get the nonlinear part of future data. Next, compare the model performance of HGFMC (1, m) and HGFMC (1, m) combined with the BP neural network.

As shown in Table 7, except for the slight decrease and no change in the forecast accuracy of SZ000796 and SH600586, the forecast accuracy of other stocks has been significantly improved.

Table 7. Comparison of MAE of out-sample forecast sequence.

Stock code	HFGMC (1, m) combined BP	HFGMC (1, m)
SZ000573	0.12	0.64
SZ000796	0.42	0.37
SZ002567	0.20	3.66
SZ002607	0.78	1.12
SZ002770	0.03	0.89
SH600586	0.21	0.21

| Show Table

DownLoad: CSV

6.7. Forecast result

This paper chose FGMC (1, m)-LLE-BP, long short-term memory neural network (LSTM), BP neural network (BP), ARIMA-BP ^[34] as the comparison model. BPNN and LSTMNN used 10 lagging data of the stock closing price to input the neural network in the form of a vector, and output the current value. For example, the input and output samples of SH600586 are shown in Table 8.

Table 8. Input and output samples of SH600586.

Sample	x₁	x₂	$\cdots$	x₉	x₁₀	y
1	2.94	2.98	$\cdots$	3.05	3.00	2.98
2	2.98	3.00	$\cdots$	3.00	2.98	2.99
			$\vdots$

| Show Table

DownLoad: CSV

The vector x in Table 8 represents the input vector, and y represents the output. This paper used a sliding window method to obtain all training samples. The forecast results of each model are shown in Figure 7.

Figure 7. Forecast results.

DownLoad: Full-Size Img PowerPoint

From the analysis of the prediction results in Figure 7, it is concluded that both SZ000573, SZ002770 and SH600586 with a small change in stock trend or SZ000796, SZ002567 and SZ002607 with a large change in stock trend predicted by the proposed model are very close to the true value. However, there are some errors between the predicted results of the comparison models and the true values. The specific model performance analysis is given in the error analysis section.

6.8. Error analysis

The forecast error analysis in this paper used mean absolute error (MAE), root mean square error (RMSE) and mean absolute percentage error (MAPE) as evaluation indicators.

$MAE = \frac{1}{n}\sum\limits_{i = 1}^n {\left| {\hat x\left( i \right) - x\left( i \right)} \right|}$

(30)

$RMSE = \sqrt {\frac{1}{n}\sum\limits_{i = 1}^n {{{\left( {\hat x\left( i \right) - x\left( i \right)} \right)}^2}} }$

(31)

$MAPE = \frac{1}{n}\sum\limits_{i = 1}^n {\left| {\frac{{\hat x\left( i \right) - x\left( i \right)}}{{x\left( i \right)}}} \right|} \times 100\%$

(32)

where $\hat x\left( i \right)$ is the predicted value, and $x\left( i \right)$ is the true value.

From the analysis of Tables 9–14, it can be seen that when the model proposed in this paper predicted SZ000573, SZ002770 and SH600586 with small changes in trend, the error between the predicted value and the true value is on average below 0.21. When predicted SZ000796, SZ002567 and SZ002607 with large changes in trend, the average error between the predicted value and the true value is less than 0.78. The performance is stable.

Table 9. Error comparison of SZ000573.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.12	0.66	0.37	0.31	0.21
RMSE	0.13	0.68	0.38	0.34	0.23
MAPE	4.3%	23%	13.2%	11%	7.4%

| Show Table

DownLoad: CSV

Table 10. Error comparison of SZ000796.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.37	1.01	1.29	1.53	1.72
RMSE	0.48	1.30	1.38	1.73	1.87
MAPE	4.5%	12.5%	16.3%	19.0%	21.6%

| Show Table

DownLoad: CSV

Table 11. Error comparison of SZ002567.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.20	0.91	0.51	0.65	0.91
RMSE	0.22	1.01	0.60	0.78	0.90
MAPE	2.6%	11.6%	6.5%	8.0 %	11.4%

| Show Table

DownLoad: CSV

Table 12. Error comparison of SZ002607.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.78	3.36	3.23	3.79	1. 90
RMSE	0.86	3.81	3.39	3.93	1.95
MAPE	4.1%	17.6%	17.1%	20.0%	10.3%

| Show Table

DownLoad: CSV

Table 13. Error comparison of SZ002770.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.03	0.05	0.16	0.19	0.14
RMSE	0.04	0.06	0.22	0.22	0.25
MAPE	1.7%	2.4%	7.6%	9.0%	6.9%

| Show Table

DownLoad: CSV

Table 14. Error comparison of SH600586.

Indicator	Proposed model	FGMC (1, m)-LLE-BP	LSTM	BP	ARIMA-BP
MAE	0.21	0.49	0.86	0.90	0.48
RMSE	0.28	0.52	0.89	0. 97	0.51
MAPE	6.9%	15.5%	26.6%	28.0%	15.1%

| Show Table

DownLoad: CSV

When FGMC (1, m)-LLE-BP predicted SZ000573, SZ002770, and SH600586 with small changes in trend, the error between the predicted value and the true value is on average below 0.66. When predicted SZ000796, SZ002567 and SZ002607 with large changes in trend, the average errors are 1.01, 0.91 and 3.36 respectively. The performance is not stable.

When LSTM neural network model predicted the SZ000573, SZ002770, and SH600586 with small trend change, the average errors are 0.37, 0.16 and 0.49 respectively. When predicted the SZ000796, SZ002567 and SZ002607 with large changes in trend, the average errors are 1.29, 0.51 and 3.23 respectively. This shows that LSTMNN has volatility in stocks with large changes in predicted trends and stocks with small changes in predicted trends.

When BP neural network model predicted the SZ000573, SZ002770, and SH600586 with small trend change, the average errors are 0.31, 0.19 and 0. 90 respectively. When predicted the SZ000796, SZ002567 and SZ002607 with large changes in trend, the average errors are 1.53, 0.65 and 3.79 respectively.

When ARIMA-BP predicted the SZ000573, SZ002770, and SH600586 with small trend change, the average errors are 0.21, 0.14 and 0. 48 respectively. When predicted the SZ000796, SZ002567 and SZ002607 with large changes in trend, the average errors are 1.72, 0.91 and 1. 90 respectively.

Combining the analysis of three error indicators, the model proposed in this paper performs best, followed by ARIMA-BP and FGMC (1, m)-LLE-BP. However, neither the LSTM nor the BP neural network model of a single prediction model can achieve the desired effect. This shows that when predicting time series with linear and non-linear factors, the combined model can achieve good results.

From Figure 8, it can be seen that the model proposed in this paper has stable and excellent performance for 6 stocks with different trends, while the performance of the other 4 comparison models is only good in some stocks. Combining the analysis of three evaluation indicators, the model proposed in this paper performed best on the MAE, RMSE, and MAPE indicators when predicted SZ002607, SZ002567, SZ000796, SH600586, SZ002770 and SZ000573. Besides, the model proposed in this paper performs better than FGMC (1, m)-LLE-BP, which shows that the HFGMC (1, m) has a certain improvement in the accuracy of prediction.

Figure 8. Error comparison of 6 stocks.

DownLoad: Full-Size Img PowerPoint

6.9. Limitations and future works

In the discussion in Section 6.2, it is pointed out that a model that deals with high-dimensional data requires certain prior data. However, it is very difficult to obtain real-time a priori data for forecasting objects with very high real-time requirements such as stock data. Generally, the high-dimensional grey model obtains the prior data by staggering the time axis of the prediction object (closing price) and prior data (indicators) (as shown in Figure 9). Therefore, the practice of staggering the time axis determines that the model cannot be used for long-term forecasting. When the time axis is staggered longer, the error may increase significantly. In addition to staggering the time axis, the use of sliding time windows to obtain prior data is also widely used in deep learning time series forecasting ^[2,3,30] (as shown in Table 8). In addition, the price of individual stocks (the experimental case of this paper) is different from stock indexes (Dow Jones, Nasdaq, Shanghai Composite Index, etc.). The price of individual stocks is usually predicted for about 10 days or 10 units of time ^[4,30,35], and the stock index reflects the general laws of the market, so long-term forecasts are feasible ^{[36,37,38,39]}. However, whether it is individual stock prices or stock indexes, the problem of prior data cannot be solved when using high-dimensional models. This is a limitation of high-dimensional models.

Figure 9. The relationship between the sequence to be predicted and the prior data.

DownLoad: Full-Size Img PowerPoint

Since directly staggering the time axis may lose the information of various indicators in the time axis, the next step can consider using the maximum entropy model to integrate all the information in the time axis as the prior information of the prediction object. Through the above work, the accuracy of the model may be more accurate.

7. Conclusions

Because the stock price is affected by many factors, the stock market can be regarded as a non-linear dynamic and grey system under the action of many factors. It is very difficult to predict stock price. There are many time series forecasting methods, but many methods do not consider that the stock market has long-term memory.

According to the characteristics of the stock market, the HFGMC (1, m)-LLE-BP is proposed. The HFGMC (1, m) is based on the FGMC (1, m) model, and combined the Hausdorff fractional derivative and Newton-Cotes formula. The HFGMC (1, m) model is used to solve the linear component in the prediction sequence. Then, the BP neural network is used to extract the non-linear component in the prediction sequence. Besides, in order to compress the redundant information in high-dimensional data, the HFGMC (1, m) model combined the locally linear embedding algorithm to reduce the dimensionality of factors that affect the stock price. Finally, the linear component and the non-linear component are added to obtain the combined prediction sequence. In the experimental part, the model proposed in this paper predicts the closing price of 6 stocks with different trends in the next 10 days. Based on the above experimental results and discussion, the following conclusions are drawn:

(i) FGMC (1, m) model has some shortcomings. Hausdorff fractional derivative and Newton-Cotes formula can improve FGMC (1, m) model. The HFGMC (1, m) model improves the prediction accuracy of time series.

(ii) Because the stock price is affected by a variety of linear and nonlinear factors, a single time prediction model cannot accurately predict the stock price. A combined forecasting model has advantages in forecasting time series affected by complex factors.

(iii) The model proposed in this paper is very stable and accurate in predicting the stock price with different historical trends. Results show that the model proposed in this paper can be widely used in the prediction of simple or complex time series.

Acknowledgments

This work is supported by National Nature Science Foundation under Grant 61862062, 61104035.

Conflict of interest

The authors declared that they have no conflict of interest.

References

[1]	Babalola OO (2010) Beneficial bacteria of agricultural importance. Biotechnol Lett 32: 1559–1570. doi: 10.1007/s10529-010-0347-0
[2]	Saleem M, Arshad M, Hussain S, et al. (2007) Perspective of plant growth promoting rhizobacteria (PGPR) containing ACC deaminase in stress agriculture. J Ind Microbiol Biotechnol 34: 635–648. doi: 10.1007/s10295-007-0240-6
[3]	Glick BR (2012) Plant growth-promoting bacteria: mechanisms and applications. Scientifica 2012: 1–15.
[4]	van der Heijden MGA, Bruin S de, Luckerhoff L, et al. (2016) A widespread plant-fungal-bacterial symbiosis promotes plant biodiversity, plant nutrition and seedling recruitment. ISME J 10: 389–399. doi: 10.1038/ismej.2015.120
[5]	Zak DR, Homes WE, White DC, et al. (2003) Plant diversity, soil microbial communities, and ecosystem function: are there any links? Ecology 84: 2042–2050. doi: 10.1890/02-0433
[6]	Wolfe BE, Klironomos JN (2005) Breaking new ground: soil communities and exotic plant invasion. Bioscience 55: 477. doi: 10.1641/0006-3568(2005)055[0477:BNGSCA]2.0.CO;2
[7]	Bais HP, Weir TL, Perry LG, et al. (2006) The role of root exudates in rhizosphere interactions with plants and other organisms. Annu Rev Plant Biol 57: 233–266. doi: 10.1146/annurev.arplant.57.032905.105159
[8]	van der Heijden MGA, Klironomos JN, Ursic M, et al. (1998) Mycorrhizal fungal diversity determines plant biodiversity, ecosystem variability and productivity. Nature 396: 69–72. doi: 10.1038/23932
[9]	Castillo CG, Rubio R, Rouanet JL, et al. (2006) Early effects of tillage and crop rotation on arbuscular mycorrhizal fungal propagules in an Ultisol. Biol Fertil Soils 43: 83–92. doi: 10.1007/s00374-005-0067-0
[10]	Fuchslueger L, Bahn M, Fritz K, et al. (2014) Experimental drought reduces the transfer of recently fixed plant carbon to soil microbes and alters the bacterial community composition in a mountain meadow. New Phytol 201: 916–927. doi: 10.1111/nph.12569
[11]	Kaiser C, Kilburn MR, Clode PL, et al. (2015) Exploring the transfer of recent plant photosynthates to soil microbes: mycorrhizal pathway vs direct root exudation. New Phytol 205: 1537–1551. doi: 10.1111/nph.13138
[12]	Behie SW, Bidochka MJ (2014) Nutrient transfer in plant-fungal symbioses. Trends Plant Sci 19: 734–740.
[13]	DeAngelis KM, Lindow SE, Firestone MK (2008) Bacterial quorum sensing and nitrogen cycling in rhizosphere soil. FEMS Microbiol Ecol 66: 197–207. doi: 10.1111/j.1574-6941.2008.00550.x
[14]	Saleem M, Moe LA (2014) Multitrophic microbial interactions for eco- and agro-biotechnological processes: theory and practice. Trends Biotechnol 32: 529–537.
[15]	Philippot L, Raaijmakers JM, Lemanceau P, et al. (2013) Going back to the roots: the microbial ecology of the rhizosphere. Nat Rev Microbiol 11: 789–799. doi: 10.1038/nrmicro3109
[16]	Fray RG (2002) Altering plant-microbe interaction through artificially manipulating bacterial quorum sensing. Ann Bot 89: 245–253. doi: 10.1093/aob/mcf039
[17]	Ishaq SL, Johnson SP, Miller ZJ, et al. (2016) Impact of cropping systems, soil inoculum, and plant species identity on soil bacterial community structure. Microb Ecol: 1–18.
[18]	Peiffer JA, Spor A, Koren O, et al. (2013) Diversity and heritability of the maize rhizosphere microbiome under field conditions. Proc Natl Acad Sci USA 110: 6548–6553. doi: 10.1073/pnas.1302837110
[19]	Chaudhry V, Rehman A, Mishra A, et al. (2012) Changes in bacterial community structure of agricultural land due to long-term organic and chemical amendments. Microb Ecol 64: 450–460. doi: 10.1007/s00248-012-0025-y
[20]	Li R, Khafipour E, Krause DO, et al. (2012) Pyrosequencing reveals the influence of organic and conventional farming systems on bacterial communities. PLoS One 7: e51897. doi: 10.1371/journal.pone.0051897
[21]	Aislabie J, Deslippe JR (2013) Soil microbes and their contribution to soil services, In: Dymond JR, Editor, Ecosystem sevices in New Zealand-conditions and trends, Lincoln: Manaaki Whenua Press, 143–161.
[22]	Lennon JT, Aanderud ZT, Lehmkuhl BK, et al. (2012) Mapping the niche space of soil microorganisms using taxonomy and traits. Ecology 93: 1867–1879. doi: 10.1890/11-1745.1
[23]	Fierer N, Bradford MA, Jackson RB (2007) Toward an ecological classification of soil bacteria. Ecology 88: 1354–1364. doi: 10.1890/05-1839
[24]	Koyama A, Wallenstein MD, Simpson RT, et al. (2014) Soil bacterial community composition altered by increased nutrient availability in Arctic tundra soils. Front Microbiol 5: 516.
[25]	Fierer N, Ladau J, Clemente JC, et al. (2013) Reconstructing the microbial diversity and function of pre-agricultural tallgrass prairie soils in the United States. Science 342: 621–624. doi: 10.1126/science.1243768
[26]	Prober SM, Leff JW, Bates ST, et al. (2015) Plant diversity predicts beta but not alpha diversity of soil microbes across grasslands worldwide. Ecol Lett 18: 85–95. doi: 10.1111/ele.12381
[27]	Tiemann LK, Grandy AS, Atkinson EE, et al. (2015) Crop rotational diversity enhances belowground communities and functions in an agroecosystem. Ecol Lett 18: 761–771. doi: 10.1111/ele.12453
[28]	Schnitzer SA, Klironomos JN, HilleRisLambers J, et al. (2011) Soil microbes drive the classic plant diversity-productivity pattern. Ecology 92: 296–303.
[29]	Barrios E (2007) Soil biota, ecosystem services and land productivity. Ecol Econ 64: 269–285. doi: 10.1016/j.ecolecon.2007.03.004
[30]	Atlas RM, Horowitz A, Krichevsky M, et al. (1991) Response of microbial populations to environmental disturbance. Microb Ecol 22: 249–256. doi: 10.1007/BF02540227
[31]	Kuan HL, Fenwick C, Glover LA, et al. (2006) Functional resilience of microbial communities from perturbed upland grassland soils to further persistent or transient stresses. Soil Biol Biochem 38: 2300–2306. doi: 10.1016/j.soilbio.2006.02.013
[32]	Orwin KH, Wardle DA (2004) New indices for quantifying the resistance and resilience of soil biota to exogenous disturbances. Soil Biol Biochem 36: 1907–1912. doi: 10.1016/j.soilbio.2004.04.036
[33]	Bérard A, Bouchet T, Sévenier G, et al. (2011) Resilience of soil microbial communities impacted by severe drought and high temperature in the context of Mediterranean heat waves. Eur J Soil Biol 47: 333–342. doi: 10.1016/j.ejsobi.2011.08.004
[34]	Thakur MP, Milcu A, Manning P, et al. (2015) Plant diversity drives soil microbial biomass carbon in grasslands irrespective of global environmental change factors. Glob Chang Biol 21: 4076–4085. doi: 10.1111/gcb.13011
[35]	Johnson SP, Miller Z, Lehnhoff EA, et al. (2017) Cropping systems modify soil biota effects on wheat (Triticum aestivum) growth and competitive ability. Weed Res 57: 6–15. doi: 10.1111/wre.12231
[36]	Somova LA, Pechurkin NS, Sarangova AB, at al. (2001) Effect of bacterial population density on germination wheat seeds and dynamics of simple artificial ecosystems. Adv Space Res 27: 1611–1615. doi: 10.1016/S0273-1177(01)00257-5
[37]	Patten CL, Glick BR (2002) Role of Pseudomonasputida indoleacetic acid in development of the host plant root system. Appl Environ Microbiol 68: 3795–3801. doi: 10.1128/AEM.68.8.3795-3801.2002
[38]	Parsons CV, Harris DMM, Patten CL (2015) Regulation of indole-3-acetic acid biosynthesis by branched-chain amino acids in Enterobacter cloacae UW5. FEMS Microbiol Lett 362: fnv153. doi: 10.1093/femsle/fnv153
[39]	Spaepen S, Vanderleyden J (2011) Auxin and plant-microbe interactions. Cold Spring Harb Perspect Biol 3: a001438.
[40]	Arkhipova TN, Veselov SU, Melentiev AI, et al. (2005) Ability of bacterium Bacillus subtilis to produce cytokinins and to influence the growth and endogenous hormone content of lettuce plants. Plant Soil 272: 201–209. doi: 10.1007/s11104-004-5047-x
[41]	Singh RP, Bijo AJ, Baghel RS, et al. (2011) Role of bacterial isolates in enhancing the bud induction in the industrially important red alga Gracilaria dura. FEMS Microbiol Ecol 76: 381–392. doi: 10.1111/j.1574-6941.2011.01057.x
[42]	Zambryski P, Tempe J, Schell J (1989) Transfer and function of T-DNA genes from agrobacterium Ti and Ri plasmids in plants. Cell 56: 193–201. doi: 10.1016/0092-8674(89)90892-1
[43]	Hooykaas PJJ, Beijersbergen AGM (1994) The virulence system of Agrobacterium tumefaciens. Annu Rev Phytopathol 32: 157–181. doi: 10.1146/annurev.py.32.090194.001105
[44]	Van de Poel B, Van Der Straeten D (2014) 1-aminocyclopropane-1-carboxylic acid (ACC) in plants: more than just the precursor of ethylene! Front Plant Sci 5: 640.
[45]	Considine PJ, Flynn N, Patching JW (1977) Ethylene production by soil microorganisms. Appl Environ Microbiol 33: 977–999.
[46]	Digiacomo F, Girelli G, Aor B, et al. (2014) Ethylene-producing bacteria that ripen fruit. ACS Synth Biol 3: 935–938. doi: 10.1021/sb5000077
[47]	Glick BR (2014) Bacteria with ACC deaminase can promote plant growth and help to feed the world. Microbiol Res 169: 30–39. doi: 10.1016/j.micres.2013.09.009
[48]	Chang P, Gerhardt KE, Huang XD, et al. (2014) Plant growth-promoting bacteria facilitate the growth of barley and oats in salt-impacted soil: implications for phytoremediation of saline soils. Int J Phytoremediation 16: 1133–1147. doi: 10.1080/15226514.2013.821447
[49]	Flematti GR, Merritt DJ, Piggott MJ, et al. (2011) Burning vegetation produces cyanohydrins that liberate cyanide and stimulate seed germination. Nat Commun 2: 360. doi: 10.1038/ncomms1356
[50]	Tao JJ, Chen HW, Ma B, et al. (2015) The role of ethylene in plants under salinity stress. Front Plant Sci 6: 1059.
[51]	Cao YR, Chen SY, Zhang JS (2008) Ethylene signaling regulates salt stress response: An overview. Plant Signal Behav 3: 761–763. doi: 10.4161/psb.3.10.5934
[52]	Fray RG, Throup JP, Daykin M, et al. (1999) Plants genetically modified to produce N-acylhomoserine lactones communicate with bacteria. Nat Biotechnol 17: 1017–1020. doi: 10.1038/13717
[53]	Barriuso J, Ramos Solano B, Fray RG, et al. (2008) Transgenic tomato plants alter quorum sensing in plant growth-promoting rhizobacteria. Plant BiotechnolJ 6: 442–452. doi: 10.1111/j.1467-7652.2008.00331.x
[54]	Rodriguez RJ, White Jr JF, Arnold AE, et al. (2009) Fungal endophytes: diversity and functional roles. New Phytol 182: 314–330. doi: 10.1111/j.1469-8137.2009.02773.x
[55]	Nair DN, Padmavathy S (2014) Impact of endophytic microorganisms on plants, environment and humans. Sci World J 2014: 250693.
[56]	Saleem M, Meckes N, Pervaiz ZH, et al. (2017) Microbial interactions in the phyllosphere increase plant performance under herbivore biotic stress. Front Microbiol 8: 41.
[57]	Vriens K, Cammue B, Thevissen K (2014) Antifungal plant defensins: mechanisms of action and production. Molecules 19: 12280–12303. doi: 10.3390/molecules190812280
[58]	Poon IK, Baxter AA, Lay FT, et al. (2014) Phosphoinositide-mediated oligomerization of a defensin induces cell lysis. Elife 3: e01808.
[59]	Wilmes M, Cammue BPA, Sahl HG, et al. (2011) Antibiotic activities of host defense peptides: more to it than lipid bilayer perturbation. Nat Prod Rep 28: 1350. doi: 10.1039/c1np00022e
[60]	Jakobsen ST (1993) Interaction between plant nutrients: IV. Interaction between calcium and phosphate. Acta Agric Scand Sect B-S P 43: 6–10.
[61]	Hepler PK (2005) Calcium: a central regulator of plant growth and development. Plant Cell 17: 2142–2155. doi: 10.1105/tpc.105.032508
[62]	Moscatiello R, Alberghini S, Squartini A, et al. (2009) Evidence for calcium-mediated perception of plant symbiotic signals in aequorin-expressing Mesorhizobium loti. BMC Microbiol 9: 206. doi: 10.1186/1471-2180-9-206
[63]	Großhans H, Filipowicz W (2008) Molecular biology: The expanding world of small RNAs. Nature 451: 414–416. doi: 10.1038/451414a
[64]	Weiberg A, Wang M, Lin FM, et al. (2013) Fungal small RNAs suppress plant immunity by hijacking host RNA interference pathways. Science 342: 118–123.
[65]	Luo L (2012) Plant cytokine or phytocytokine. Plant Signal Behav 7: 1513–1514. doi: 10.4161/psb.22425
[66]	Saleem M, Ji H, Amirullah A, et al. (2017) Pseudomonas syringae pv. tomato DC3000 growth in multiple gene knockouts predicts interactions among hormonal, biotic and abiotic stress responses. Eur J Plant Pathol: 1–8.
[67]	Koch A, Biedenkopf D, Furch A, et al. (2016) An RNAi-based control of Fusarium graminearum infections through spraying of long dsRNAs involves a plant passage and is controlled by the fungal silencing machinery. PLoS Pathog 12: e1005901. doi: 10.1371/journal.ppat.1005901
[68]	Shakeri J, Foster HA (2007) Proteolytic activity and antibiotic production by Trichoderma harzianum in relation to pathogenicity to insects. Enzyme Microb Technol 40: 961–968. doi: 10.1016/j.enzmictec.2006.07.041
[69]	Ghisalberti EL, Sivasithamparam K (1991) Antifungal antibioitcs produced by Trichoderma spp. Soil Biol Biochem 23: 1011–1020. doi: 10.1016/0038-0717(91)90036-J
[70]	Ji SH, Paul NC, Deng JX, et al. (2013) Biocontrol activity of Bacillus amyloliquefaciens CNU114001 against fungal plant diseases. Mycobiology 41: 234–242. doi: 10.5941/MYCO.2013.41.4.234
[71]	Yu SM (2009) Biological control of postharvest green and blue mold rots of citrus fruits by Bacillus amyloliquefaciens BCL251. Daejeon: Chungnam National University.
[72]	Lee DG (2010) Biological control of strawberry anthracnose using endophytic bacteria, Bacillus amyloliquefaciens CP1. Daejeon: Chungnam National University.
[73]	Yu GY, Sinclair JB, Hartman GL, et al. (2002) Production of iturin A by Bacillus amyloliquefaciens suppressing Rhizoctonia solani.Soil Biol Biochem 34: 955–963. doi: 10.1016/S0038-0717(02)00027-5
[74]	Lee JH, Seo MW, Kim HG (2012) Isolation and characterization of an antagonistic endophytic bacterium Bacillus velezensis CB3 the control of citrus green mold pathogen Penicillium digitatum. Korean J Mycol 40: 118–123. doi: 10.4489/KJM.2012.40.2.118
[75]	Santhanam R, Luu VT, Weinhold A, et al. (2005) Native root-associated bacteria rescue a plant from a sudden-wilt disease that emerged during continuous cropping. Proc Natl Acad Sci USA 112: E5013–E5020.
[76]	Bainton NJ, Stead P, Chhabra SR, et al. (1992) N-(3-Oxohexanoyl)-L-homoserine lactone regulates carbapenem antibiotic production in Erwinia carotovora. Biochem J 288: 997–1004. doi: 10.1042/bj2880997
[77]	Jones S, Yu B, Bainton NJ, et al. (1993) The lux autoinducer regulates the production of exoenzyme virulence determinants in Erwinia carotovora and Pseudomonas aeruginosa. EMBO J 12: 2477–2482.
[78]	Chernin LS, Winson MK, Thompson JM, et al. (1998) Chitinolytic activity in Chromobacterium violaceum: substrate analysis and regulation by quorum sensing. J Bacteriol 180: 4435–4441.
[79]	Wood DW, Pierson LS (1996) The phzI gene of Pseudomonas aureofaciens 30–84 is responsible for the production of a diffusible signal required for phenazine antibiotic production. Gene 168: 49–53. doi: 10.1016/0378-1119(95)00754-7
[80]	Carruthers FL (1994) A study of antifungal activity by a potential biological control strain, Pseudomonas aureofaciens strain PA147-2. University of Canterbury.
[81]	Gamard P, Sauriol F, Benhamou N, et al. (1997) Novel butyrolactones with antifungal activity produced by Pseudomonas aureofaciens strain 63-28. J Antibiot 50: 742–749. doi: 10.7164/antibiotics.50.742
[82]	Paulitz T, Nowak-Thompson B, Gamard P, et al. (2000) A novel antifungal furanone from Pseudomonas aureofaciens, a biocontrol agent of fungal plant pathogens. J Chem Ecol 26: 1515–1524. doi: 10.1023/A:1005595927521
[83]	Vincent MN, Harrison LA, Brackin JM, et al. (1991) Genetic analysis of the antifungal activity of a soilborne Pseudomonasaureofaciens strain. Appl Environ Microbiol 57: 2928–2934.
[84]	Agrios GN (2005) Plant pathology. Elsevier Academic Press, 922.
[85]	Benhamou N, Belanger RR, Paulitz TC (1996) Pre-inoculation of Ri T-DNA-transformed pea roots with Pseudomonas fluorescens inhibits colonization by Pythium ultimum Trow: an ultrastructural and cytochemical study. Planta 199: 105–117.
[86]	Askary H, Benhamou N, Brodeur J (1997) Ultrastructural and cytochemical investigations of the antagonistic effect of Verticillium lecanii on cucumber powdery mildew. Phytopathology 87: 359–368. doi: 10.1094/PHYTO.1997.87.3.359
[87]	Nobbe F, Hiltner L (1896) Inoculation of the soil for cultivating leguminous plants (patent).
[88]	Shuang S, Zhenfang G, Xiaolei G (2015) The effect of bacteria on seed germination in sorghum and rape under cadmium and petroleum conditions. Int J Biotechnol Wellness Ind 4: 123–127.
[89]	Harper SHT, Lynch JM (1980) Microbial effects on the germination and seedling growth of barley. New Phytol 84: 473–481. doi: 10.1111/j.1469-8137.1980.tb04554.x
[90]	Perry G (2014) Ethylene induced soil microbes to increase seed germination, reduce growth time, and improve crop yield in Pisum sativum L. Peer J.
[91]	Babalola OO, Berner DK, Amusa NA (2007) Evaluation of some bacterial isolates as germination stimulants of Striga hermonthica.African J Agric Res 2: 27–30.
[92]	Ye J, Kostrzynska M, Dunfield K, et al. (2010) Control of Salmonella on sprouting mung bean and alfalfa seeds by using a biocontrol preparation based on antagonistic bacteria and lytic bacteriophages. J Food Prot 73: 9. doi: 10.4315/0362-028X-73.1.9
[93]	Broadfoot M (2016) Microbes added to seeds could boost crop production. Sci Am.
[94]	Greenfield LG, Mcmanus MT, Outred HA, et al. (2000) The microbial decomposition of seeds. Agron Soc New Zeal: 47–51.
[95]	Pitty A, Staniforth DW, Tiffany LH (1987) Fungi associated with caryopses of Setaria species from field-harvested seeds and from soil under two tillage systems. WeedSci 35: 319–323.
[96]	Hatfield JL, Buhler DD, Stewart BA (1998) Integrated weed and soil management, Ann Arbor: Sleeping Bear Press, 21.
[97]	Ganley RJ, Newcombe G (2006) Fungal endophytes in seeds and needles of Pinus monticola. Mycol Res 110: 318–327. doi: 10.1016/j.mycres.2005.10.005
[98]	Saikkonen K, Ion D, Gyllenberg M (2002) The persistence of vertically transmitted fungi in grass metapopulations. Proc R Soc B Biol Sci 269: 1397–1403. doi: 10.1098/rspb.2002.2006
[99]	Tintjer T, Leuchtmann A, Clay K, et al. (2008) Variation in horizontal and vertical transmission of the endophyte Epichloë elymi infecting the grass Elymus hystrix. New Phytol 179: 236–246. doi: 10.1111/j.1469-8137.2008.02441.x
[100]	Pollnac FW, Maxwell BD, Menalled FD (2009) Using species-area curves to examine weed communities in organic and conventional spring wheat systems. Weed Sci 57: 241–247. doi: 10.1614/WS-08-159.1
[101]	Pershina E, Valkonen J, Kurki P, et al. (2015) Comparative analysis of prokaryotic communities associated with organic and conventional farming systems. PLoS One 10: e0145072. doi: 10.1371/journal.pone.0145072
[102]	Flohre A, Rudnick M, Traser G, et al. (2011) Does soil biota benefit from organic farming in complex vs. simple landscapes? Agric Ecosyst Environ 141: 210–214.
[103]	Hartmann M, Frey B, Mayer J, et al. (2015) Distinct soil microbial diversity under long-term organic and conventional farming. ISME J 9: 1177–1194. doi: 10.1038/ismej.2014.210
[104]	Luo P, Han X, Wang Y, et al. (2015) Influence of long-term fertilization on soil microbial biomass, dehydrogenase activity, and bacterial and fungal community structure in a brown soil of northeast China. Ann Microbiol 65: 533–542. doi: 10.1007/s13213-014-0889-9
[105]	Ramirez KS, Lauber CL, Knight R, et al. (2010) Consistent effects of nitrogen fertilization on soil bacterial communities in contrasting systems. Ecology 91: 3463–3470. doi: 10.1890/10-0426.1
[106]	Zhalnina K, Dias R, de Quadros PD, et al. (2015) Soil pH determines microbial diversity and composition in the park grass experiment. Microb Ecol 69: 395–406. doi: 10.1007/s00248-014-0530-2
[107]	Campbell BJ, Polson SW, Hanson TE, et al. (2010) The effect of nutrient deposition on bacterial communities in Arctic tundra soil. Environ Microbiol 12: 1842–1854. doi: 10.1111/j.1462-2920.2010.02189.x
[108]	Su JQ, Ding LJ, Xue K, et al. (2015) Long-term balanced fertilization increases the soil microbial functional diversity in a phosphorus-limited paddy soil. Mol Ecol 24: 136–150. doi: 10.1111/mec.13010
[109]	Rousk J, Bååth E, Brookes PC, et al. (2010) Soil bacterial and fungal communities across a pH gradient in an arable soil. ISME J 4: 1340–1351. doi: 10.1038/ismej.2010.58
[110]	Lauber CL, Hamady M, Knight R, et al. (2009) Pyrosequencing-based assessment of soil pH as a predictor of soil bacterial community structure at the continental scale. Appl Environ Microbiol 75: 5111–5120. doi: 10.1128/AEM.00335-09
[111]	Roesch LFW, Fulthorpe RR, Riva A, et al. (2007) Pyrosequencing enumerates and contrasts soil microbial diversity. ISME J 1: 283–290.
[112]	McKenzie SC, Goosey HB, O'Neill KM, et al. (2016) Integration of sheep grazing for cover crop termination into market gardens: Agronomic consequences of an ecologically based management strategy. Renew Agric Food Syst 74: 1–14.
[113]	Liu N, Kan HM, Yang GW, et al. (2015) Changes in plant, soil, and microbes in a typical steppe from simulated grazing: explaining potential change in soil C. Ecol Monogr 85: 269–286. doi: 10.1890/14-1368.1
[114]	Liu N, Zhang Y, Chang S, et al. (2012) Impact of grazing on soil carbon and microbial biomass in typical steppe and desert steppe of Inner Mongolia. PLoS One 7: e36434. doi: 10.1371/journal.pone.0036434
[115]	Taddese G, Peden D, Hailemariam A, et al. (2007) Effect of livestock grazing on soil microorganisms of cracking and self mulching vertisol. Ethopain Vet J 11: 141–150.
[116]	Mofidi M, Rashtbari M, Abbaspour H, et al. (2012) Impact of grazing on chemical, physical and biological properties of soils in the mountain rangelands of Sahand, Iran. Rangel J 34: 297. doi: 10.1071/RJ11087
[117]	Chen W, Huang D, Liu N, et al. (2015) Improved grazing management may increase soil carbon sequestration in temperate steppe. Sci Rep 5: 10892. doi: 10.1038/srep10892
[118]	Duzy LM, Price AJ, Balkcom KS, et al. (2016) Assessing the economic impact of inversion tillage, cover crops, and herbicide regimes in Palmer Amaranth (Amaranthus palmeri) infested cotton. Int J Agron 2016: 1–9.
[119]	Adusumilli N, Fromme D (2016) Evaluating benefits and costs of cover crops in cotton production systems in Northwestern Louisiana. Southern Agricultural Economics Association, 2016 Annual Meeting.
[120]	Sulc RM, Franzluebbers AJ (2014) Exploring integrated crop-livestock systems in different ecoregions of the United States. Eur J Agron 57: 21–30. doi: 10.1016/j.eja.2013.10.007
[121]	Dabney SM, Delgado JA, Reeves DW (2001) Using winter cover crops to improve soil and water quality. Commun Soil Sci Plant Anal 32: 1221–1250. doi: 10.1081/CSS-100104110
[122]	Gallandt ER, Liebman M, Huggins DR (1998) Improving soil quality: implications for weed management. J Crop Prod 2: 95–121.
[123]	Dabney SM, Schreiber JD, Rothrock CS, et al. (1996) Cover crops affect sorghum seedling growth. Agron J 88: 961. doi: 10.2134/agronj1996.00021962003600060020x
[124]	Liebman M, Davis AS (2000) Integration of soil, crop and weed management in low-external-input farming systems. Weed Res 40: 27–47. doi: 10.1046/j.1365-3180.2000.00164.x
[125]	Reeves DW (1994) Cover crops and rotations, In: Hatfield JL, Stewart BA, Editors, Advances in Soil Science: Crop Residue Management, Boca Raton: Lewis Publishers, 125–172.
[126]	Hartwig NL, Ammon HU (2002) Cover crops and living mulches. Weed Sci 50: 688–699. doi: 10.1614/0043-1745(2002)050[0688:AIACCA]2.0.CO;2
[127]	Kamh M, Horst WJ, Amer F, et al. (1999) Mobilization of soil and fertilizer phosphate by cover crops. Plant Soil 211: 19–27. doi: 10.1023/A:1004543716488
[128]	Wild A (1993) Soils and the environment. Cambridge: Cambridge University Press.
[129]	Hu SJ, van Bruggen AHC, Grünwald NJ (1999) Dynamics of bacterial populations in relation to carbon availability in a residue-amended soil. Appl Soil Ecol 13: 21–30. doi: 10.1016/S0929-1393(99)00015-3
[130]	Snapp SS, Swinton SM, Labarta R, et al. (2004) Evaluating cover crops for benefits, costs and performance within cropping system niches. Agron J 97: 322–332.
[131]	Ghimire R, Norton JB, Stahl PD, et al. (2014) Soil microbial substrate properties and microbial community responses under irrigated organic and reduced-tillage crop and forage production systems. PLoS One 9: e103901. doi: 10.1371/journal.pone.0103901
[132]	Biederbeck VO, Zentner RP, Campbell CA (2005) Soil microbial populations and activities as influenced by legume green fallow in a semiarid climate. Soil Biol Biochem 37: 1775–1784. doi: 10.1016/j.soilbio.2005.02.011
[133]	Grayston SJ, Wang SQ, Campbell CD, et al. (1998) Selective influence of plant species on microbial diversity in the rhizosphere. Soil Biol Biochem 30: 369–378. doi: 10.1016/S0038-0717(97)00124-7
[134]	McGonigle TP, Evans DG, Miller MH (1990) Effect of degree of soil disturbance on mycorrhizal colonization and phosphorus absorption by maize in growth chamber and field experiments. New Phytol 116: 629–636. doi: 10.1111/j.1469-8137.1990.tb00548.x
[135]	Mathew RP, Feng Y, Githinji L, et al. (2012) Impact of no-tillage and conventional tillage systems on soil microbial communities. Appl Environ Soil Sci 2012: 10.
[136]	García-Orenes F, Morugán-Coronado A, Zornoza R, et al. (2013) Changes in soil microbial community structure influenced by agricultural management practices in a mediterranean agro-ecosystem. PLoS One 8: e80522. doi: 10.1371/journal.pone.0080522
[137]	Lupwayi NZ, Rice WA, Clayton GW (1998) Soil microbial diversity and community structure under wheat as influenced by tillage and crop rotation. Soil Biol Biochem 30: 1733–1741. doi: 10.1016/S0038-0717(98)00025-X
[138]	De Quadros PD, Zhalnina K, Davis-Richardson A, et al. (2012) The effect of tillage system and crop rotation on soil microbial diversity and composition in a subtropical acrisol. Diversity 4: 375–395. doi: 10.3390/d4040375
[139]	Brevik E (2013) The potential impact of climate change on soil properties and processes and corresponding influence on food security. Agriculture 3: 398–417. doi: 10.3390/agriculture3030398
[140]	el Fantroussi S, Verschuere L, Verstraete W, et al. (1999) Effect of phenylurea herbicides on soil microbial communities estimated by analysis of 16S rRNA gene fingerprints and community-level physiological profiles. Appl Environ Microbiol 65: 982–988.
[141]	Lupwayi NZ, Harker KN, Clayton GW, et al. (2004) Soil microbial biomass and diversity after herbicide application. Can J Plant Sci 84: 677–685. doi: 10.4141/P03-121
[142]	Lo CC (2010) Effect of pesticides on soil microbial community. J Environ Sci Health B 45: 348–359.
[143]	Hussain S, Siddique T, Saleem M, et al. (2009) Chapter 5 impact of pesticides on soil microbial diversity, enzymes, and biochemical reactions. Adv Agron 102: 159–200. doi: 10.1016/S0065-2113(09)01005-0
[144]	Das N, Chandran P (2011) Microbial degradation of petroleum hydrocarbon contaminants: an overview. Biotechnol Res Int 2011: 941810.
[145]	Kuhad RC, Johri AK, Singh A, et al. (2004) Bioremediation of pesticide-contaminated soils, In: Singh A, Ward OP, Editors, Applied Bioremediation and Phytoremediation, Springer Berlin Heidelberg, 35–54.
[146]	Hussain S, Arshad M, Saleem M, et al. (2007) Screening of soil fungi for in vitro degradation of endosulfan. World J Microbiol Biotechnol 23: 939–945. doi: 10.1007/s11274-006-9317-z
[147]	Hussain S, Arshad M, Saleem M, et al. (2007) Biodegradation of α- and β-endosulfan by soil bacteria. Biodegradation 18: 731–740. doi: 10.1007/s10532-007-9102-1
[148]	Hussain S, Siddique T, Arshad M, et al. (2009) Bioremediation and phytoremediation of pesticides: recent advances. Crit Rev Environ Sci Technol 39: 843–907. doi: 10.1080/10643380801910090
[149]	Singh DK (2008) Biodegradation and bioremediation of pesticide in soil: concept, method and recent developments. Indian J Microbiol 48: 35–40. doi: 10.1007/s12088-008-0004-7
[150]	Kumar M, Philip L (2006) Enrichment and isolation of a mixed bacterial culture for complete mineralization of endosulfan. J Environ Sci Heal Part B 41: 81–96. doi: 10.1080/03601230500234935
[151]	Kumar M, Philip L (2006) Bioremediation of endosulfan contaminated soil and water-Optimization of operating conditions in laboratory scale reactors. J Hazard Mater 136: 354–364. doi: 10.1016/j.jhazmat.2005.12.023
[152]	Baczynski TP, Pleissner D, Grotenhuis T (2010) Anaerobic biodegradation of organochlorine pesticides in contaminated soil-Significance of temperature and availability. Chemosphere 78: 22–28. doi: 10.1016/j.chemosphere.2009.09.058
[153]	Langenhoff AAM, Staps JJM, Pijls C, et al. (2002) Intrinsic and stimulated in situ biodegradation of hexachlorocyclohexane (HCH). Water, Air Soil Pollut Focus 2: 171–181. doi: 10.1023/A:1019943410568
[154]	Lamichhane KM, Babcock Jr RW, Turnbull SJ, et al. (2012) Molasses enhanced phyto and bioremediation reatability study of explosives contaminated Hawaiian soils. J Hazard Mater 243: 334–339. doi: 10.1016/j.jhazmat.2012.10.043
[155]	Hussain A, Iqbal Quzi J, Abdullah Shakir H (2014) Implication of molasses as electron donor for biological sulphate reduction. Am J Environ Eng 4: 7–10.
[156]	Antonious GF (2012) On-farm bioremediation of dimethazone and trifluralin residues in runoff water from an agricultural field. J Environ Sci Heal Part B 47: 608–621. doi: 10.1080/03601234.2012.668454
[157]	Ma L, Deng F, Yang C, et al. (2016) Bioremediation of PAH-contaminated farmland: field experiment. Environ Sci Pollut Res: 1–9.
[158]	Du RY, Wen D, Zhao PH, et al. (2016) Effect of bacterial application on metal availability and plant growth in farmland-contaminated soils. J Bioremediation Biodegrad 7: 341.
[159]	Azizul M, Omine K (2013) Bioremediation of agricultural land damaged by tsunami, In: Chamy R, Editors, Biodegradation of Hazardous Special Products, Intech.
[160]	Nabti E, Schmid M, Hartmann A (2015) Application of halotolerant bacteria to restore plant growth under salt stress, In: Maheshwari D, Saraf M, Editors, Halophiles. Springer International Publishing, 235–259.
[161]	Fierer N, Lauber CL, Ramirez KS, et al. (2012) Comparative metagenomic, phylogenetic and physiological analyses of soil microbial communities across nitrogen gradients. ISME J 6: 1007–1017. doi: 10.1038/ismej.2011.159
[162]	Greening C, Carere CR, Rushton-Green R, et al. (2015) Persistence of the dominant soil phylum Acidobacteria by trace gas scavenging. Proc Natl Acad Sci USA 112: 10497–10502. doi: 10.1073/pnas.1508385112
[163]	Kielak AM, Barreto CC, Kowalchuk GA, et al. (2016) The ecology of Acidobacteria: moving beyond genes and genomes. Front Microbiol 7: 744.
[164]	Robinson CH, Wookey PA, Parsons AN, et al. (1995) Responses of plant litter decomposition and nitrogen mineralisation to simulated environmental change in a high arctic polar semi-desert and a subarctic dwarf shrub heath. Oikos 74: 503. doi: 10.2307/3545996
[165]	McHale PJ, Mitchell MJ, Bowles FP (1998) Soil warming in a northern hardwood forest: trace gas fluxes and leaf litter decomposition. Can J For Res 28: 1365–1372. doi: 10.1139/x98-118
[166]	Luo Y, Wan S, Hui D, Wallace LL (2001) Acclimatization of soil respiration to warming in a tall grass prairie. Nature 413: 622–625. doi: 10.1038/35098065
[167]	Myers SS, Zanobetti A, Kloog I, et al. (2014) Increasing CO2 threatens human nutrition. Nature 510: 139–142. doi: 10.1038/nature13179
[168]	Dietterich LH, Zanobetti A, Kloog I, et al. (2015) Impacts of elevated atmospheric CO2 on nutrient content of important food crops. Sci Data 2: 150036. doi: 10.1038/sdata.2015.36
[169]	De La Puente LS, Perez PP, Martinez-Carrasco R, et al. (2000) Action of elevated CO2 and high temperatures on the mineral chemical composition of two varieties of wheat. Agrochimica 44: 506.
[170]	Seneweera SP, Conroy JP (1997) Growth, grain yield and quality of rice (Oryza sativa L.) in response to elevated CO2 and phosphorus nutrition, In: Ando T, et al. Editors, Plant Nutrition for Sustainable Food Production and Environment, Dordrecht: Springer, 873–878
[171]	Poorter H, Van Berkel Y, Baxter R, et al. (1997) The effect of elevated CO2 on the chemical composition and construction costs of leaves of 27 C3 species. Plant, Cell Environ 20: 472–482. doi: 10.1046/j.1365-3040.1997.d01-84.x
[172]	Caldwell CR, Britz SJ, Mirecki RM (2005) Effect of temperature, elevated carbon dioxide, and drought during seed development on the isoflavone content of dwarf soybean [Glycine max (L.) Merrill] grown in controlled environments. J Agric Food Chem 53: 1125–1129.
[173]	Reddy KR, Zhao D (2005) Interactive effects of elevated CO2 and potassium deficiency on photsynthesis, growth, and biomass partitioning in cotton. F Crop Res 94: 201–213. doi: 10.1016/j.fcr.2005.01.004
[174]	Barcenas-Moreno G, Gomez-Brandon M, Rousk J, et al. (2009) Adaptation of soil microbial communities to temperature: comparison of fungi and bacteria in a laboratory experiment. Glob Chang Biol 15: 2950–2957.
[175]	Alster CJ, Koyama A, Johnson NG, et al. (2016) Temperature sensitivity of soil microbial communities: An application of macromolecular rate theory to microbial respiration. JGeophys Res Biogeosciences 121: 1420–1433. doi: 10.1002/2016JG003343
[176]	Schindlbacher A, Rodler A, Kuffner M, et al. (2011) Experimental warming effects on the microbial community of a temperate mountain forest soil. Soil Biol Biochem 43: 1417–1425. doi: 10.1016/j.soilbio.2011.03.005
[177]	Pietikainen J, Pettersson M, Baath E (2005) Comparison of temperature effects on soil respiration and bacterial and fungal growth rates. FEMS Microbiol Ecol 52: 49–58. doi: 10.1016/j.femsec.2004.10.002
[178]	Curtin D, Beare MH, Hernandez-Ramirez G (2012) Temperature and moisture effects on microbial biomass and soil organic matter mineralization. Soil Sci Soc Am J 76: 2055. doi: 10.2136/sssaj2012.0011
[179]	Biederbeck VO, Campbell CA (1973) Soil microbial activity as influenced by temperature trends and fluctations. Can J Soil Sci 53: 363–376. doi: 10.4141/cjss73-053
[180]	Auffret MD, Karhu K, Khachane A, et al. (2016) The role of microbial community composition in controlling soil respiration responses to temperature. PLoS One 11: e0165448. doi: 10.1371/journal.pone.0165448
[181]	Richardson RE, James CA, Bhupathiraju VK, et al. (2002) Microbial activity in soils following steam trreatment. Biodegradation 13: 285–295. doi: 10.1023/A:1021257026932
[182]	Brussaard L, de Ruiter PC, Brown GG (2007) Soil biodiversity for agricultural sustainability. Agric Ecosyst Environ 121: 233–244. doi: 10.1016/j.agee.2006.12.013

This article has been cited by:

1.	Yuer Yang, Ruotong Du, Haodong Tang, Yanxin Zheng, 2021, SSLPNet: A financial econometric prediction model for small-sample long panel data, 9781450384971, 174, 10.1145/3512576.3512607
2.	Xiao Ren, Jie Hua, Xin Chi, Yao Tan, Visual analysis of social events and stock market volatility in China and the USA during the pandemic, 2022, 20, 1551-0018, 1229, 10.3934/mbe.2023056
3.	Yuzhen Chen, Hui Wang, Suzhen Li, Rui Dong, A Novel Grey Seasonal Model for Natural Gas Production Forecasting, 2023, 7, 2504-3110, 422, 10.3390/fractalfract7060422
4.	Wanli Xie, Zhenguo Xu, Caixia Liu, Jianyue Chen, A novel fractional Hausdorff grey system model and its applications, 2023, 45, 10641246, 7575, 10.3233/JIFS-230121
5.	Yunfei Yang, Xiaomei Wang, Jiamei Xiong, Lifeng Wu, Yifang Zhang, An innovative method for short-term forecasting of blockchain cryptocurrency price, 2025, 138, 0307904X, 115795, 10.1016/j.apm.2024.115795
6.	Xiaojun Zhou, Chunna Zhao, Yaqun Huang, Chengli Zhou, Junjie Ye, Improved fractional-order gradient descent method based on multilayer perceptron, 2025, 183, 08936080, 106970, 10.1016/j.neunet.2024.106970
7.	Gazi Murat Duman, Elif Kongar, A novel Hausdorff fractional grey Bernoulli model and its Application in forecasting electronic waste, 2025, 3, 29497507, 349, 10.1016/j.wmb.2025.02.002
8.	Yuansheng Qian, Zhijie Zhu, Xinsong Niu, Linyue Zhang, Kang Wang, Jianzhou Wang, Environmental policy-driven electricity consumption prediction: A novel buffer-corrected Hausdorff fractional grey model informed by two-stage enhanced multi-objective optimization, 2025, 377, 03014797, 124540, 10.1016/j.jenvman.2025.124540

Reader Comments

Your name:*

Email:*
© 2017 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)

通讯作者: 陈斌, bchen63@163.com

1.
沈阳化工大学材料科学与工程学院沈阳 110142

AIMS Microbiology

2.7 7.0

Metrics

Article views(9317) PDF downloads(1833) Cited by(27)

Preview PDF

Download XML

Export Citation

Article outline

Show full outline

AIMS Microbiology

Plant-microbial interactions in agriculture and the use of farming systems to improve diversity and productivity

Related Papers:

Abstract

1. Introduction

2. Specific frameworks of the proposed model

3. The FGMC (1, m) model

4. The HFGMC (1, m)

4.1. Derivation of Hausdorff Fractional Accumulate Generation Operation (HFAGO)

4.2. Optimization of background value based on Newton-Cotes formula

5. Local linear embedding algorithm and BP neural network

5.1. Local linear embedding (LLE) dimensionality reduction algorithm

5.2. BP neural network

6. Experiment analysis

6.1. Experimental examples

6.2. Discussion of hyperparameters

6.3. Discussion on the stability of stock prices

6.4. Select parameters

6.5. Discussion between HFGMC (1, m) and FGMC (1, m)

6.6. Discussion of BP Neural Network

6.7. Forecast result

6.8. Error analysis

6.9. Limitations and future works

7. Conclusions

Acknowledgments

Conflict of interest

References

This article has been cited by:

Reader Comments

通讯作者: 陈斌, bchen63@163.com

Metrics

Other Articles By Authors

Catalog

AIMS Microbiology

Plant-microbial interactions in agriculture and the use of farming systems to improve diversity and productivity

Related Papers:

Abstract

1. Introduction

2. Specific frameworks of the proposed model

3. The FGMC (1, m) model

4. The HFGMC (1, m)

4.1. Derivation of Hausdorff Fractional Accumulate Generation Operation (HFAGO)

4.2. Optimization of background value based on Newton-Cotes formula

5. Local linear embedding algorithm and BP neural network

5.1. Local linear embedding (LLE) dimensionality reduction algorithm

5.2. BP neural network

6. Experiment analysis

6.1. Experimental examples

6.2. Discussion of hyperparameters

6.3. Discussion on the stability of stock prices

6.4. Select parameters

6.5. Discussion between HFGMC (1, m) and FGMC (1, m)

6.6. Discussion of BP Neural Network

6.7. Forecast result

6.8. Error analysis

6.9. Limitations and future works

7. Conclusions

Acknowledgments

Conflict of interest

References

This article has been cited by:

Reader Comments

通讯作者: 陈斌, bchen63@163.com

Metrics

Other Articles By Authors

Related pages

Tools

Export File

Citation

Format

Content

Catalog