Remaining useful life prediction of the lithium-ion battery based on CNN-LSTM fusion model and grey relational analysis

Dewang Chen; Xiaoyu Zheng; Ciyang Chen; Wendi Zhao; Dewang Chen; Xiaoyu Zheng; Ciyang Chen; Wendi Zhao

doi:10.3934/era.2023031

Electronic Research Archive

2023, Volume 31, Issue 2: 633-655. doi: 10.3934/era.2023031

Previous Article Next Article

Research article Special Issues

Remaining useful life prediction of the lithium-ion battery based on CNN-LSTM fusion model and grey relational analysis

1.
School of Transportation, Fujian University of Technology, Fuzhou 350118, China
2.
State Key Laboratory of Management and Control for Complex Systems, Institute of Automation, Chinese Academy of Sciences, Beijing 100190, China
3.
Fujian Rural Credit Union, Fuzhou 350003, China

Received: 08 October 2022 Revised: 30 October 2022 Accepted: 03 November 2022 Published: 15 November 2022

The performance of lithium-ion batteries will decline dramatically with the increase in usage time, which will cause anxiety in using lithium-ion batteries. Some data-driven models have been employed to predict the remaining useful life (RUL) model of lithium-ion batteries. However, there are limitations to the accuracy and applicability of traditional machine learning models or just a single deep learning model. This paper presents a fusion model based on convolutional neural network (CNN) and long short-term memory network (LSTM), named CNN-LSTM, to measure the RUL of lithium-ion batteries. Firstly, this model uses the grey relational analysis to extract the main features affecting the RUL as the health index (HI) of the battery. In addition, the fusion model can capture the non-linear characteristics and time-space relationships well, which helps find the capacity decay and failure threshold of lithium-ion batteries. The experimental results show that: 1) Traditional machine learning is less effective than LSTM. 2) The CNN-LSTM fusion model is superior to the single LSTM model in predicting performance. 3) The proposed model is superior to other comparable models in error indexes, which could reach 0.36% and 0.38e-4 in mean absolute percentage error (MAPE) and mean square error (MSE), respectively. 4) The proposed model can accurately find the failure threshold and the decay fluctuation for the lithium-ion battery.

Keywords:

Citation: Dewang Chen, Xiaoyu Zheng, Ciyang Chen, Wendi Zhao. Remaining useful life prediction of the lithium-ion battery based on CNN-LSTM fusion model and grey relational analysis[J]. Electronic Research Archive, 2023, 31(2): 633-655. doi: 10.3934/era.2023031

Related Papers:

[1]	Xiaoling Chen, Xingfa Zhang, Yuan Li, Qiang Xiong . Daily LGARCH model estimation using high frequency data. Data Science in Finance and Economics, 2021, 1(2): 165-179. doi: 10.3934/DSFE.2021009
[2]	Paarth Thadani . Financial forecasting using stochastic models: reference from multi-commodity exchange of India. Data Science in Finance and Economics, 2021, 1(3): 196-214. doi: 10.3934/DSFE.2021011
[3]	Kexian Zhang, Min Hong . Forecasting crude oil price using LSTM neural networks. Data Science in Finance and Economics, 2022, 2(3): 163-180. doi: 10.3934/DSFE.2022008
[4]	Moses Khumalo, Hopolang Mashele, Modisane Seitshiro . Quantification of the stock market value at risk by using FIAPARCH, HYGARCH and FIGARCH models. Data Science in Finance and Economics, 2023, 3(4): 380-400. doi: 10.3934/DSFE.2023022
[5]	Samuel Asante Gyamerah, Collins Abaitey . Modelling and forecasting the volatility of bitcoin futures: the role of distributional assumption in GARCH models. Data Science in Finance and Economics, 2022, 2(3): 321-334. doi: 10.3934/DSFE.2022016
[6]	Alejandro Rodriguez Dominguez, Om Hari Yadav . A causal interactions indicator between two time series using extreme variations in the first eigenvalue of lagged correlation matrices. Data Science in Finance and Economics, 2024, 4(3): 422-445. doi: 10.3934/DSFE.2024018
[7]	Wojciech Kurylek . Are Natural Language Processing methods applicable to EPS forecasting in Poland?. Data Science in Finance and Economics, 2025, 5(1): 35-52. doi: 10.3934/DSFE.2025003
[8]	Nitesha Dwarika . The risk-return relationship in South Africa: tail optimization of the GARCH-M approach. Data Science in Finance and Economics, 2022, 2(4): 391-415. doi: 10.3934/DSFE.2022020
[9]	Mohamed F. Abd El-Aal . Analysis Factors Affecting Egyptian Inflation Based on Machine Learning Algorithms. Data Science in Finance and Economics, 2023, 3(3): 285-304. doi: 10.3934/DSFE.2023017
[10]	Xiaozheng Lin, Meiqing Wang, Choi-Hong Lai . A modification term for Black-Scholes model based on discrepancy calibrated with real market data. Data Science in Finance and Economics, 2021, 1(4): 313-326. doi: 10.3934/DSFE.2021017

Abstract

1. Introduction

Estimating foreign exchange rate (FX) volatility is a core risk management activity for financial institutions, corporates and regulators. The subject has been extensively investigated among both practitioners and scientific researchers, and several alternative models exist. Among the most prominent are the models belonging to GARCH and stochastic volatility classes. However, the true value of volatility cannot be directly observed. Hence, volatility must be estimated, inevitably with error. This constitutes a fundamental problem in implementing parametric models, especially in the context of high-frequency data. Andersen and Bollerslev (1998) proposed using realized volatility, as derived from high-frequency data, to accurately measure the true latent integrated volatility. This approach has gained attention for volatility modeling in markets where tick-level data is available (Andersen et al., 2013). Andersen et al. (2003) suggest fractionally integrated ARFIMA models in this context. Still, the long-memory HAR (heterogeneous autoregressive) model of is arguably the most widely used to capture the high persistence typically observed in realized volatility of financial prices. The HAR model is relatively simple and easy to estimate. In empirical applications, the model tends to perform better than GARCH and stochastic volatility models possibly due to the sensitivity of tightly parameterized volatility models to minor model misspecifications (). Although realized volatility $(RV)$ is a consistent estimator of the true latent volatility, it is subject to measurement error in empirical finite samples. Hence, $RV$ will not only reflect the true latent integrated volatility $(IV)$ , but also additional measurement errors. Bollerslev et al. (2016) propose utilizing higher-order realized moments of the realized distribution to approximate these measurement errors. More specifically, Bollerslev et al. (2016) propose the HARQ-model, which augments the HAR model with realized quarticity as an additional covariate.

The empirical performance of the HARQ and related extensions has been extensively studied. The focus has predominantly been on equity markets. A majority of the studies analyze U.S. data; see Bollerslev et al. (2016); Clements, A. and Preve, D. (2021); Pascalau and Poirier (2023); Andersen et al. (2023) and others. Liu et al. (2018) and Wang et al. (2020) investigate Chinese equity markets, whereas Liang et al. (2022); Ma et al. (2019) analyse international data. Bitcoin and electricity markets have attracted some attention; see, for instance, Shen et al. (2020); Qieu et al. (2021), and Qu et al. (2018).

Empirical applications of the HARQ model in the context of foreign exchange rate risk are sparse. Lyócsa et al. (2016) find that the standard HAR model rarely is outperformed by less parsimonious specifications on CZKEUR, PLZEUR, and HUFEUR data. Plíhal et al. (2021) and Rokicka and Kudła (2020) estimate the HARQ model on EURUSD and EURGBP data, respectively. Their focus is different from ours, as they investigate the incremental predictive power of implied volatility for a broad class of HAR models. In a similar vein, Götz (2023) and Lyócsa et al. (2024) utilize the HARQ model for the purpose of estimating foreign exchange rate tail risk.

Using updated tick-level data from two major currency pairs, EURUSD and USDJPY, this paper documents the relevance of realized quarticity for improving volatility estimates across varying forecasting horizons. These results are robust across estimation windows, evaluation metrics, and model specifications.

2. Materials and method

2.1. Data

We use high-frequency intraday ticklevel spot data, publicly available at DukasCopy^* The sample period is 1. January 2010 to 31. December 2022. Liu et al. (2015) investigate the optimal intraday sampling frequency across a significant number of asset classes and find that 5-min intervals usually outperform others. Hence, as common in the literature, we estimate the realized volatility from 5-minute returns.

^*This data source is also used by Plíhal et al. (2021), Risstad et al. (2023) and Lyócsa et al. (2024), among others.

To filter tick-level data, we follow a two-step cleaning procedure based on the recommendations by Barndorff-Nielsen et al. (2009). Initially, we eliminate data entries that exhibit any of the following issues: (i) absence of quotes, (ii) a negative bid-ask spread, (iii) a bid-ask spread exceeding 50 times the median spread of the day, or (iv) a mid-quote deviation beyond ten mean absolute deviations from a centered mean (computed excluding the current observation from a window of 25 observations before and after). Following this, we calculate the mid-quotes as the average of the bid and ask quotes and then resample the data at 5-minute intervals.

We compute the consistent estimator of the true latent time- $t$ variance from

$\begin{equation} RV^{2}_{t} \equiv \sum\limits^{M}_{t = 1} r^{2}_{t, i}, \end{equation}$

(1)

where $M = 1/\Delta$ , and the $\Delta$ -period intraday return is $r_{t, i} \equiv \log(S_{t-1+i\times\Delta}) - \log(S_{t-1+(i-1)\times\Delta})$ , where $S$ is the spot exchange rate. Analogously, the multi $(h)$ -period realized variance estimator is

$\begin{equation} RV^{2}_{t-1, t-h} = \frac{1}{h} \sum\limits_{i = 1}^{h}RV^{2}_{t-h}. \end{equation}$

(2)

Setting $h = 5$ and $h = 22$ yields weekly and monthly estimates, respectively.

Table 1 displays descriptive statistics for daily realized variances, as computed from (1).

Table 1. Realized Variance (daily).

	Min	Mean	Median	Max	$\rho_1$
EURUSD	0.1746	3.0606	2.2832	59.4513	0.5529
USDJPY	0.1018	3.2460	2.0096	168.0264	0.2860
The table contains summary statistics for the daily $R V \mathrm{~s}$ for EURUSD and USDJPY. $\rho_1$ is the standard first order autocorrelation coefficient. Sample period: 1. January 2010 to 31. December 2022.

| Show Table

DownLoad: CSV

2.2. The HARQ models

To represent the long-memory dynamic dependencies in volatility, Corsi (2009) proposed using daily, weekly, and monthly lags of realized volatility as covariates. The original HAR model is defined as

$\begin{equation} R V_{t} = \beta_{0}+\beta_{1} R V_{t-1}+\beta_{2} R V_{t-1 \mid t-5}+\beta_{3} R V_{t-1 \mid t-22}+u_{t}, \end{equation}$

(3)

where $RV$ is computed from (1) and (2). If the variables in (2.2) contain measurement errors, the beta coefficients will be affected. Bollerslev et al. (2016) suggests two measures to alleviate this. First, they include a proxy for measurement error as an additional explanatory variable. Furthermore, they directly adjust the coefficients in proportion to the magnitude of the measurement errors:

$\begin{aligned} R V_{t} = & \beta_{0}+\underbrace{\left(\beta_{1}+\beta_{1 Q} R Q_{t-1}^{1 / 2}\right)}_{\beta_{1, t}} R V_{t-1} \\ & +\underbrace{\left(\beta_{2}+\beta_{2 Q} R Q_{t-1 \mid t-5}^{1 / 2}\right)}_{\beta_{2, t}} R V_{t-1 \mid t-5} \\ & +\underbrace{\left(\beta_{3}+\beta_{3 Q} R Q_{t-1 \mid t-22}^{1 / 2}\right)}_{\beta_{3, t}} R V_{t-1 \mid t-22}+u_{t}, \end{aligned}$

where realized quarticity $RQ$ is defined as

$\begin{equation} R Q_{t} \equiv \frac{M}{3} \sum\limits_{i = 1}^{M} r_{t, i}^{4} \end{equation}$

(4)

The full HARQ model in (2.2) adjusts the coefficients on all lags of $RV$ . A reasonable conjecture is that measurement errors in realized volatilities tend to diminish at longer forecast horizons, as these errors are diversified over time. This suggests that measurement errors in daily lagged realized volatilities are likely to be relatively more important. Motivated by this Bollerslev et al. (2016) specify the HARQ model as

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\underbrace{\left(\beta_{1}+\beta_{1 Q} R Q_{t-1}^{1 / 2}\right)}_{\beta_{1, t}} R V_{t-1}+\beta_{2} R V_{t-1 \mid t-5} \\ & +\beta_{3} R V_{t-1 \mid t-22}+u_{t} . \end{aligned} \end{equation}$

(5)

Although there is no reason to expect that autoregressive models of order one will be able to accurately capture long memory in realized volatility, we estimate AR(1) models as a point of reference. The AR and ARQ models are defined as

$\begin{equation} \begin{aligned} R V_{t} = \beta_{0}+\beta_{1} RV_{t-1}+u_{t}. \end{aligned} \end{equation}$

(6)

and

$\begin{equation} \begin{aligned} R V_{t} = \beta_{0}+\underbrace{\left(\beta_{1}+\beta_{1 Q} R Q_{t-1}^{1 / 2}\right)}_{\beta_{1, t}} R V_{t-1}+u_{t}. \end{aligned} \end{equation}$

(7)

in equations (6) and (7), respectively.

3. Results and discussion

Due to noisy data and related estimation errors, forecasts from realized volatility models might occasionally appear as unreasonably high or low. Thus, in line with Swanson et al. (1997) and Bollerslev et al. (2016), we filter forecasts from all models so that any forecast outside the empirical distribution of the estimation sample is replaced by the sample mean.

3.1. In-sample estimation results

reports in-sample parameter estimates for the ARQ, HARQ, and HARQ-F models, along with the benchmark AR and ARQ models, for one-day ahead EURUSD (upper panel) and USDJPY (lower panel) volatility forecasts. Robust standard errors (s.e.) are computed as proposed by . $R^{2}$ , $MSE$ , and $QLIKE$ are displayed at the bottom of each panel.

Table 2. In-sample estimation results, one-day-ahead volatility forecasts.

EURUSD	AR	HAR	ARQ	HARQ	HARQ-F
$\beta_{0}$	$1.3663^*$	$0.3961^*$	$0.7428^*$	$0.2785^*$	$-$ 0.0651
s.e.	0.1843	0.0598	0.0969	0.0586	0.0685
$\beta_{1}$	$0.5530^{*}$	$0.2364^{*}$	$0.7903^{*}$	$0.4349^{*}$	$0.3740^{*}$
s.e.	0.0653	0.0730	0.0388	0.0754	0.0792
$\beta_{2}$		$0.3767^{*}$		$0.3072^{*}$	$0.4613^{*}$
s.e.		0.0717		0.0697	0.1031
$\beta_{3}$		$0.2572^{*}$		$0.1850^{*}$	$0.2398^{*}$
s.e.		0.0532		0.0515	0.0822
$\beta_{1 Q}$			$-2.4914^{*}$	$-1.3708^{*}$	$-0.9710^{*}$
s.e.			0.3377	0.1939	0.2266
$\beta_{2 Q}$					$-1.7578^{***}$
s.e.					0.8706
$\beta_{3 Q}$					$-3.9819^{*}$
s.e.					1.1618
$R^{2}$	0.3058	0.3956	0.3685	0.4101	0.4166
$MSE$	6.3005	5.4852	5.7315	5.3538	5.2950
$QLIKE$	0.1647	0.1230	0.1540	0.1217	0.1199
USDJPY	AR	HAR	ARQ	HARQ	HARQ-F
$\beta_{0}$	$2.3073^{*}$	$1.0682^{*}$	$1.3537^{*}$	$0.7811^{*}$	$0.5218^{*}$
s.e.	0.2362	0.1381	0.2207	0.1429	0.1328
$\beta_{1}$	$0.2854^{*}$	$0.1819^{*'}$	$0.6180^{*}$	$0.5177^{***}$	$0.4416^{*}$
s.e.	0.0804	0.0806	0.0853	0.1106	0.1260
$\beta_{2}$		$0.1441^{**}$		0.0542	$0.2345^{***}$
s.e.		0.0585		0.0543	0.1072
$\beta_{3}$		$0.3443^{*}$		$0.2188^{*}$	$0.2228^{*}$
s.e.		0.0499		0.0493	0.0658
$\beta_{1 Q}$			$-0.2295^{*}$	$-0.1967^{*}$	$-0.1526^{*}$
s.e.			0.0318	0.0386	0.0476
$\beta_{2 Q}$					$-0.2296^{**}$
s.e.					0.0849
$\beta_{3 Q}$					$-0.3573^{*}$
s.e.					0.1142
$R^{2}$	0.0814	0.1154	0.1489	0.1581	0.1642
$MSE$	33.6096	32.3668	31.1409	30.8063	30.5818
$QLIKE$	0.3214	0.2561	0.2663	0.2377	0.2242
Note: The table contains in-sample parameter estimates and corresponding standard errors (), together with $R^{2}$ . $MSE$ and $QLIKE$ computed from (12) and (13). Superscripts , , and ** represent statistical significance in a two-sided t-test at 1%, 5% and 10% levels, respectively.

| Show Table

DownLoad: CSV

The coefficients $\beta_{1 Q}$ are negative and exhibit strong statistical significance, aligning with the hypothesis that $RQ$ represents time-varying measurement error. When comparing the autoregressive (AR) coefficient of the AR model to the autoregressive parameters in the ARQ model, the AR coefficient is markedly lower, reflecting the difference in in persistence between the models.

In the comparative analysis of the HAR and HARQ models applied to both currency pairs, the HAR model assigns more emphasis to the weekly and monthly lags, which are generally less sensitive to measurement errors. In contrast, the HARQ model typically assigns a higher weight to the daily lag. However, when measurement errors are substantial, the HARQ model reduces the weight on the daily lag to accommodate the time-varying nature of the measurement errors in the daily realized volatility ( $RV$ ). The flexible version of this model, the HARQ-F, allows for variability in the weekly and monthly lags, resulting in slightly altered parameters compared to the standard HARQ model. Notably, the coefficients $\beta_{2 Q}$ and $\beta_{3 Q}$ in the HARQ-F model are statistically significant, and this model demonstrates a modest enhancement in in-sample fit relative to the HARQ model.

3.2. Out-of-sample forecasting results

To further assess the out-of-sample performance of the HARQ model, we consider three alternative HAR type specifications. More specifically, we include both the HAR-with-Jumps (HAR-J) and the Continuous-HAR (CHAR) proposed by Andersen et al. (2007), as well as the SHAR model proposed by Patton and Sheppard (2015), in the forecasting comparisons. Based on the Bi-Power Variation (BPV) measure of Barndorff-Nielsen and Shephard (2004), HAR-J and CHAR decompose the total variation into a continuous and a discontinuous (jump) part.

The HAR-J model augments the standard HAR model with a measure of the jump variation;

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\beta_{1} R V_{t-1}+\beta_{2} R V_{t-1 \mid t-5}+\beta_{3} R V_{t-1 \mid t-22} \\ & +\beta_{J} J_{t-1}+u_{t}, \end{aligned} \end{equation}$

(8)

where $J_{t} \equiv \max \left[R V_{t}-B P V_{t}, 0\right]$ , and the $B P V$ measure is defined as,

$\begin{equation} BPV_{t} \equiv \mu_{1}^{-2} \sum\limits_{i = 1}^{M-1}\left|r_{t, i}\right|\left|r_{t, i+1}\right|, \end{equation}$

(9)

with $\mu_{1} = \sqrt{2 / \pi} = \mathbb{E}(|Z|)$ , and $Z$ is a standard normal random variable.

The CHAR model includes measures of the continuous component of the total variation as covariates;

$\begin{equation} R V_{t} = \beta_{0}+\beta_{1} B P V_{t-1}+\beta_{2} B P V_{t-1 \mid t-5}+\beta_{3} B P V_{t-1 \mid t-22}+u_{t}. \end{equation}$

(10)

Inspired by the semivariation measures of , propose the SHAR model, which, in contrast to the HAR model, effectively allows for asymmetric responses in volatility forecasts from negative and positive intraday returns. More specifically, when $R V_{t}^{-} \equiv \sum_{i = 1}^{M} r_{t, i}^{2} \mathbb{I}_{\left\{r_{t, i} < 0\right\}}$ and $R V_{t}^{+} \equiv \; \sum_{i = 1}^{M} r_{t, i}^{2} \mathbb{I}_{\left\{r_{t, i} > 0\right\}}$ , the SHAR model is defined as:

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\beta_{1}^{+} R V_{t-1}^{+}+\beta_{1}^{-} R V_{t-1}^{-}+\beta_{2} R V_{t-1 \mid t-5} \\ & +\beta_{3} R V_{t-1 \mid t-22}+u_{t} . \end{aligned} \end{equation}$

(11)

To evaluate model performance, we consider the mean squared error (MSE) and the QLIKE loss, which, according to Patton (2011), both are robust to noise. MSE is defined as

$\begin{equation} MSE(RV_{t}, F_{t}) \equiv (RV_{t}-F_{t})^2, \end{equation}$

(12)

where $F_{t}$ refers to the one-period direct forecast. QLIKE is defined as

$\begin{equation} QLIKE(RV_{t}, F_{t}) \equiv \frac{RV_{t}}{F_{t}} - \ln\left(\frac{RV_{t}}{F_{t}}\right) -1. \end{equation}$

(13)

3.2.1. Daily forecasting horizon

contains one-day-ahead forecasts for EURUSD and USDJPY. The table reports model performance, expressed as model loss normalized by the loss of the HAR model. Each row reflects a combination of estimation window and loss function. The lowest ratio on each row, highlighting the best, performing model, is in bold. We evaluate the models using both a rolling window $(RW)$ and an expanding window $(EW)$ . In both cases, forecasts are derived from model parameters re-estimated each day with a fixed length $RW$ comprised of the previous 1000 days, as well as an $EW$ using all of the available observations. The sample sizes for $EW$ thus range from 1000 to 3201 days. The results are consistent in that the HARQ-F model is the best performer for both currency pairs and across loss functions and estimation windows. The HARQ model is closest to HARQ-F. Neither HAR-J, CHAR, nor SHAR appear to consistently improve upon the standard HAR model.

Table 3. Out-of-sample forecast losses, one-day-ahead volatility forecasts.

EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.1483	1.0000	1.0088	0.9945	1.0080	1.0311	0.9759	0.9655^*
MSE-EW	1.1619	1.0000	0.9984	0.9908	1.0050	1.02660	0.9742	0.9720^*
QLIKE-RW	1.3153	1.0000	0.9907	0.9813	1.0078	1.1575	0.9767	0.9582^*
QLIKE-EW	1.3915	1.0000	0.9907	0.9944	1.0052	1.1927	0.9952	0.9721^**
USDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.0502	1.0000	1.0053	0.9979	1.0238	0.8907	0.8885	0.8832^*
MSE-EW	1.0475	1.0000	1.0243	1.0133	1.0515	0.9558	0.9446	0.9376^*
QLIKE-RW	1.2320	1.0000	1.0748	0.9944	0.9811	0.9482	0.8824	0.8667^*
QLIKE-EW	1.3066	1.0000	1.0023	0.9800	0.9941	1.0039	0.8949	0.8519^*
Note: Model performance, expressed as model loss normalized by the loss of the HAR model. Each row reflects a combination of estimation window and loss function. Ratio for the best performing model on each row in bold. Corresponding asterix * and ** denote 1% and 5% confidence levels from Diebold-Mariano test for one-sided tests of superior performance of the best performing model compared to the HAR model.

| Show Table

DownLoad: CSV

Judging from , it is beneficial to include RQ as an explanatory variable when RV is measured inaccurately. However, precise measurement of RV becomes more difficult when RV is high, inducing a positive correlation between RV and RQ. At the same time, high $RV$ often coincides with jumps. To clarify whether the performance of RQ-based models is due to jump dynamics, further segments the results in into forecasts for days when the previous day's $RQ$ was very high (Top 5% RQ, Table 4b) and the remaining sample (Bottom 95% RQ, Table 4a). As this breakdown shows, the RQ-based models perform relatively well also during periods of non-extreme heteroscedasticity of RQ.

Table 4. Stratified one-day-ahead out-of-sample forecast losses.

(a) Bottom 95% RQ
EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.1156	1.0000	0.9937	0.9907	1.0021	1.0636	0.9925	0.9794
MSE-IW	1.1175	1.0000	0.9887	0.9885	1.0020	1.0711	0.9967	0.9866
QLIKE-RW	1.3299	1.0000	0.9975	0.9855	1.0071	1.1598	0.9745	0.9555
QLIKE-IW	1.4108	1.0000	0.9956	0.9980	1.0055	1.1995	0.9944	0.9720
heightUSDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.0330	1.0000	1.0146	0.9984	0.9940	0.9592	0.9526	0.9495
MSE-IW	1.0590	1.0000	0.9962	0.9925	1.0001	0.9849	0.9681	0.9601
QLIKE-RW	1.2507	1.0000	1.1353	0.9877	0.9829	0.9542	0.8797	0.8450
QLIKE-IW	1.3266	1.0000	0.9883	0.9734	0.9993	1.0100	0.8887	0.8434
(b) Top 5% RQ
EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.2276	1.0000	1.0453	1.0036	1.0225	0.9523	0.9355	0.9316
MSE-IW	1.2642	1.0000	1.0206	0.9960	1.0121	0.9218	0.9224	0.9382
QLIKE-RW	1.0876	1.0000	0.8851	0.9152	1.0186	1.1223	1.0116	0.9996
QLIKE-IW	1.0902	1.0000	0.9141	0.9389	1.0006	1.0856	1.0081	0.9745
USDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.0674	1.0000	1.0025	0.9974	1.0535	0.9425	0.8700	0.8518
MSE-IW	1.0347	1.0000	1.0566	1.0365	1.1090	0.9246	0.9183	0.9126
QLIKE-RW	1.0202	1.0000	1.5755	1.0697	0.9601	0.8803	0.9135	0.9999
QLIKE-IW	1.0544	1.0000	1.1789	1.0628	0.9279	0.9278	0.9730	0.9588
Note: The table segments the results in according to $RQ$ . The bottom panel shows the ratios for days following a value of $RQ$ in the top $5\%$ . The top panel shows the results for the remaining $95\%$ of sample. Ratio for the best performing model on each row in bold.

| Show Table

DownLoad: CSV

3.2.2. Longer forecast horizons

In practitioner applications, longer forecasts than one day are often of interest. We now extend our analysis to weekly and monthly horizons, using direct forecasts. The daily forecast analysis in subsubsection 3.2.1 indicates the lag order of RQ plays an important role in forecast accuracy. Hence, following Bollerslev et al. (2016), we consider the HARQ-h model, and adjust the lag corresponding to the specific forecast horizon only. Specifically, for the weekly and monthly forecasts analysed here, the relevant HARQ-h specifications become

$\begin{equation} \begin{aligned} R V_{t+4 \mid t} = & \beta_{0}+\beta_{1} R V_{t-1}+\underbrace{\left(\beta_{2}+\beta_{2 Q} R Q_{t-1 \mid t-5}^{1 / 2}\right)}_{\beta_{2, t}} R V_{t-1 \mid t-5} \\ & +\beta_{3} R V_{t-1 \mid t-22}+u_{t} \end{aligned} \end{equation}$

(14)

and

$\begin{equation} \begin{aligned} R V_{t+21 \mid t} = & \beta_{0}+\beta_{1} R V_{t-1}+\beta_{2} R V_{t-1 \mid t-5}+u_{t}, \\ & +\underbrace{\left(\beta_{3}+\beta_{3 Q} R Q_{t-1 \mid t-22}^{1 / 2}\right)}_{\beta_{3, t}} R V_{t-1 \mid t-22}+u_{t}, \end{aligned} \end{equation}$

(15)

respectively.

presents in-sample parameter estimates across model specifications. The patterns observed here closely resemble those of the daily estimates detailed in . All coefficients on RQ ( $\beta_{1Q}, \beta_{2Q}, \beta_{3Q}$ ) are negative, except for the $(h = 22)$ lag statistically significant. This indicates that capturing measurement errors is relevant also for forecast horizons beyond one day. The HARQ model consistently allocates greater weight to the daily lag compared to the standard HAR model. Similarly, the HARQ-h model predominantly allocates its weight towards the time-varying lag. The weights of the HARQ-F model on the different lags are relatively more stable when compared to the HARQ-h model.

Table 5. In-sample weekly and monthly model estimates.

(a) EURUSD
	Weekly						Monthly
	AR	ARQ	HAR	HARQ	HARQ-F	HARQ-h	AR	ARQ	HAR	HARQ	HARQ-F	HARQ-h
$\beta_{0}$	$0.8646^{*}$	$0.2634^{*}$	$0.5680^{*}$	$0.4758^{*}$	-0.0250	$0.2275^{**}$	$1.6388^{*}$	$0.9642^{*}$	$0.9269^{*}$	$0.8452^{*}$	0.2328	0.2153
s.e.	0.1345	0.0927	0.0997	0.0882	0.0895	0.0861	0.1806	0.1840	0.2246	0.2099	0.2080	0.2093
$\beta_{1}$	$0.7168^{*}$	$0.9620^{*}$	$0.1194^{*}$	$0.2752^{*}$	$0.1836^{*}$	$0.1181^{*}$	$0.4616^{*}$	$0.7373^{*}$	$0.0717^{*}$	$0.2097^{*}$	$0.1131^{*}$	$0.0646^{*}$
s.e.	0.0480	0.0400	0.0264	0.0395	0.0269	0.0214	0.0564	0.0616	0.0205	0.0401	0.0248	0.0185
$\beta_{2}$			$0.3938^{*}$	$0.3395^{*}$	$0.5777^{*}$	$0.7635^{*}$			$0.2091^{*}$	$0.1606^{*}$	$0.3706^{*}$	$0.2176^{*}$
s.e.			0.0887	0.0881	0.1282	0.1139			0.0587	0.0554	0.0962	0.0563
$\beta_{3}$			$0.3008^{*}$	$0.2440^{*}$	$0.3131^{**}$	0.0876			$0.4163^{*}$	$0.3661^{*}$	$0.5153^{*}$	$0.7179^{*}$
s.e.			0.0880	0.0817	0.1275	0.0940			0.1186	0.1174	0.1498	0.1106
$\beta_{1Q}$		$-5.4876^{*}$		$-1.0749^{*}$	$-0.4728^{*}$			$-6.1534^{*}$		$-0.9499^{*}$	$-0.3246^{*}$
s.e.		0.4817		0.1377	0.1005			0.9900		0.1815	0.0846
$\beta_{2Q}$					$-2.7357^{*}$	$-4.9739^{*}$					$-2.3111^{*}$
s.e.					0.9302	0.7181					0.8020
$\beta_{3Q}$					$-5.6441^{*}$						$-7.8467^{*}$	$-10.9979^{*}$
s.e.					1.4540						2.1071	1.9082
$R^{2}$	0.5138	0.5642	0.5453	0.5604	0.5843	0.5756	0.4297	0.5191	0.5072	0.5237	0.5678	0.5568
$MSE$	2.6073	2.3370	2.4385	2.3576	2.2292	2.2759	2.1913	1.8477$	1.8932	1.8299	1.6606	1.7027
$QLIKE$	0.0862	0.0731	0.0752	0.0735	0.0679	0.0704	0.1073	0.0804	0.0839	0.0012	0.0760	0.0788
(b) USDJPY
	Weekly						Monthly
	AR	ARQ	HAR	HARQ	HARQ-F	HARQ-h	AR	ARQ	HAR	HARQ	HARQ-F	HARQ-h
$\beta_{0}$	$2.0305^{*}$	1.1976^*	$1.3310^{*}$	$1.1591^{*}$	$0.8708^{*}$	$0.9646^{*}$	$2.5786^{*}$	$2.2815^{*}$	$1.7358^{*}$	$1.6356^{*}$	$1.3894^{*}$	$1.3900^{*}$
s.e.	0.2484	0.1550	0.1967	0.1646	0.1701	0.1564	0.1928	0.2245	0.2792	0.2678	0.3106	0.3151
$\beta_{1}$	$0.3709^{*}$	$0.6801^{*}$	$0.0687^{*}$	$0.2722^{*}$	$0.1650^{*}$	$0.0668^{*}$	0.2011^*	$0.3121^{*}$	$0.0286^{**}$	$0.1460^{*}$	$0.0829^{*}$	$0.0283^{*}$
s.e.	0.0717	0.0512	0.0266	0.0500	0.0373	0.0207	0.0363	0.0566	0.0119	0.0258	0.0166	0.0113
$\beta_{2}$			0.1294	0.0742	$0.3558^{*}$	$0.4971^{*}$			$0.0865^{*}$	0.0541	0.1886^*	$0.0923^{*}$
s.e.			0.0700	0.0609	0.0787	0.0790			0.0389	0.0333	0.0487	0.0376
$\beta_{3}$			$0.3910^{*}$	$0.3147^{*}$	$0.2622^{*}$	0.1829^*			$0.3460^{*}$	$0.3030^{*}$	$0.3340^{*}$	$0.4811^{*}$
s.e.			0.0703	0.0621	0.0959	0.0693			0.0916	0.0883	0.1346	0.1220
$\beta_{1Q}$		$-0.6085^{*}$		$-0.1190^{*}$	$-0.0571^{*}$			$-0.2167^{**}$		$-0.0678^{*}$	$-0.0318^{*}$
s.e.		0.0534		0.0173	0.0141			0.0832		0.0093	0.0068
$\beta_{2Q}$					$-0.3653^{*}$	$-0.5357^{*}$					$-0.1659^{*}$
s.e.					0.0704	0.0648					0.0465
$\beta_{3Q}$					-0.2750						-0.3946	$-0.7392^{**}$
s.e.					0.2010						0.2942	0.2900
$R^{2}$	0.1367	0.2323	0.1848	0.2270	0.2557	0.2475	0.1414	0.2106	.2205	0.2496	0.2761	0.2542
$MSE$	11.6923	$ 10.3980	11.0412	10.4701	10.0811	10.1919	5.4365	4.9983	4.9351	4.7513	4.58326	4.7220
$QLIKE$	0.2361	0.4197	0.2057	0.1937	0.4076	0.1405	0.2143	0.1973	0.1801	0.1734	0.1634	0.1680
Note: In-sample parameter estimates for weekly $(h=5)$ and monthly $(h=22)$ forecasting models. EURUSD in upper panel (Table 5a) and USDJPY in lower panel (Table 5b). Robust standard errors (s.e.) using accommodate autocorrelation up to order 10 $(h=5)$ , and 44 $(h=22)$ , respectively. Superscripts , * and *** represent statistical significance in a two-sided t-test at 1%, 5%, and 10% levels.

| Show Table

DownLoad: CSV

Table 6 and Table 7 detail the out-of-sample performance for weekly and monthly forecasts, respectively. Notably, the HAR-J, CHAR, and SHAR models generally fail to demonstrate consistent improvements over the basic HAR model. This is a sharp contrast to the RQ-augmented models. The HARQ-F model outperforms the HAR model both for EURUSD and USDJPY for nearly all instances. Also, HARQ-h delivers forecasts that are relatively consistent with the HAR model. Judging from both weekly and monthly results, the inherent flexibility of the HARQ-F is beneficial also for longer-term forecasts. We note that, at the monthly forecasting horizon for USDJPY, there is some variability as to preferred Q-specifications. Also, in some monthly instances, the Diebold-Mariano null hypothesis of equal predictability cannot be rejected. This is not unreasonable, since the number of independent monthly observations naturally becomes lower than for corresponding shorter forecasting horizons, leading to higher parameter uncertainty and related noise in volatility estimates.

Table 6. Weekly out-of-sample forecast losses.

EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F	HARQ-h
MSE-RW	1.3063	1.0000	0.9636	0.9884	1.0017	1.1459	0.9677	0.9024^*	0.9205
MSE-EW	1.2702	1.0000	0.9433	0.9559	0.9997	1.1288	0.9501	0.8996^*	0.9117
QLIKE-RW	1.5923	1.0000	0.9819	0.9840	0.9995	1.3558	0.9932	0.8701	0.9283
QLIKE-EW	1.7682	1.0000	0.9874	1.0031	1.0033	1.4134	0.9648	0.8832^*	0.9297
USDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F	HARQ-h
MSE-RW	1.0618	1.0000	0.9464	0.9509	0.9965	0.9064	0.8971	0.8393^*	0.8443
MSE-EW	1.1707	1.0000	1.0148	1.0021	1.0336	1.0194	0.9388	0.8993	0.8976^*
QLIKE-RW	1.3119	1.0000	1.0057	0.9910	0.9740	1.0493	0.9099	0.8246^*	0.8359
QLIKE-EW	1.3847	1.0000	0.9918	0.9768	1.0002	1.1391	0.9179	0.8350^*	0.8463
Note: Model performance, expressed as model loss normalized by the loss of the HAR model. Each row reflects a combination of estimation window and loss function. Ratio for the best-performing model on each row in bold. Corresponding asterix * and ** denote 1% and 5% confidence levels from Diebold-Mariano test for one-sided tests of superior performance of the best performing model compared to the HAR model.

| Show Table

DownLoad: CSV

Table 7. Monthly out-of-sample forecast losses.

EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F	HARQ-h
MSE-RW	1.3289	1.0000	0.9876	0.9952	1.0003	1.1876	0.9625	0.8803^*	0.9004
MSE-IW	1.3265	1.0000	0.9759	1.0010	1.0044	1.1707	0.9537	0.8723	0.9070
QLIKE-RW	1.4301	1.0000	0.9945	0.9950	0.9982	1.2380	0.9622	0.9215^*	0.9279
QLIKE-IW	1.5155	1.0000	0.9951	1.0051	1.0011	1.2596	0.9599	0.9333	0.9784
USDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F	HARQ-h
MSE-RW	1.2529	1.0000	1.0215	1.0086	0.9893	1.5820	1.0500	1.0070	0.9621^*
MSE-IW	1.2547	1.0000	1.0073	1.0029	1.0119	1.1181	0.9620	0.9495^*	0.9780
QLIKE-RW	1.1937	1.0000	1.0023	0.9963	0.9893	1.0313	0.9307	0.9454	1.0318
QLIKE-IW	1.2894	1.0000	0.9959	0.9909	1.0000	1.1453	0.9452	0.8932^*	1.0143
Note: Model performance, expressed as model loss normalized by the loss of the HAR model. Each row reflects a combination of estimation window and loss function. Ratio for the best performing model on each row in bold. Corresponding asterix * and ** denote 1% and 5% confidence levels from Diebold-Mariano test for one-sided tests of superior performance of the best performing model compared to the HAR model.

| Show Table

DownLoad: CSV

3.3. Robustness

3.3.1. Alternative HARQ Specifications

The intention of the HARQ model is to capture the heteroskedastic measurement error of realized variance. The HARQ model in (5) approximates this through the square root of $RQ$ . argues that this encounters possible issues with numerical stability. Still, this specification is somewhat ad-hoc and a number of reasonable alternatives exist. To clarify whether the performance of the HARQ model is sensitive to the definition of $RQ$ , we follow and substitute $R Q, R Q^{-1 / 2}, R Q^{-1}$ , and $\log (R Q)$ in place of $R Q^{1/2}$ . Furthermore, we augment the standard HAR and HARQ models with $R Q^{1 / 2}$ as an additional explanatory variable, which allows the $\operatorname{HAR}(\mathrm{Q})$ model intercept to be time-varying.

reports the out-of-sample forecast results from the alternative HARQ specifications. We normalize all losses by those of the HARQ model based on $R Q^{1/2}$ .

Table 8. Alternative HARQ Specifications.

	Alternative $RQ$ transformations					Adding $RQ^{1/2}$
EURUSD	$R Q$	$R Q^{1 / 2}$	$R Q^{-1 / 2}$	$R Q^{-1}$	$\log (R Q)$	HAR	HARQ
MSE-RW	1.0023	1.0000	1.0263	1.0246	1.0092	1.0309	1.0052
MSE-IW	1.0016	1.0000	1.0274	1.0265	1.0069	1.0292	1.0086
QLIKE-RW	1.0042	1.0000	1.0326	1.0304	1.0007	1.0250	1.0067
QLIKE-IW	1.0014	1.0000	1.0064	1.0254	0.9937	1.0044	1.0164
USDJPY	$R Q$	$R Q^{1 / 2}$	$R Q^{-1 / 2}$	$R Q^{-1}$	$\log (R Q)$	HAR	HARQ
MSE-RW	1.0001	1.0000	1.1345	1.1225	1.0516	1.1202	1.0118
MSE-IW	1.0049	1.0000	1.0606	1.0543	0.9931	1.0512	1.0186
QLIKE-RW	1.0097	1.0000	1.1439	1.1067	0.9794	1.0731	1.0455
QLIKE-IW	1.0188	1.0000	1.1105	1.0841	0.9322	1.0358	0.9989
Note: Model performance, expressed as model loss normalized by the loss of the HARQ model, relies on $R Q^{1/2}$ . Each row reflects a combination of estimation window and loss function. Ratio for the best, performing model on each row in bold. The left panel reports the results based on alternative $RQ$ interaction terms. The right panel reports the results from including $R Q^{1/2}$ as an explanatory variable.

| Show Table

DownLoad: CSV

The two rightmost columns of reveal that including $R Q^{1/2}$ as an explanatory variable in the HAR and HARQ models does not lead to improved forecasts. Similarly, applying alternative RQ transformations does not appear to be helpful. Overall, we conclude that the HARQ model demonstrates greater stability and is generally favored over the alternative specifications.

3.3.2. Alternative Q-Models

HARQ is essentially an expansion of the HAR model. In a similar vein, the other benchmark volatility models can be extended accordingly. Following Bollerslev et al. (2016), from the HAR-J model defined in (3.2), we construct the HARQ-J model;

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\left(\beta_{1}+\beta_{1 Q} R Q_{t-1}^{1 / 2}\right) R V_{t-1}+\beta_{2} R V_{t-1 \mid t-5} \\ & +\beta_{3} R V_{t-1 \mid t-22}+\beta_{J} J_{t-1}+u_{t}. \end{aligned} \end{equation}$

(16)

Furthermore, from the CHAR model defined in (3.2), we construct the CHARQ model;

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\left(\beta_{1}+\beta_{1 Q} T P Q_{t-1}^{1 / 2}\right) B P V_{t-1}+\beta_{2} B P V_{t-1 \mid t-5} \\ & +\beta_{3} B P V_{t-1 \mid t-22}+u_{t}. \end{aligned} \end{equation}$

(17)

Lastly, from the SHAR model defined in (3.2), we construct the SHARQ model;

$\begin{equation} \begin{aligned} R V_{t} = & \beta_{0}+\left(\beta_{1}^{+}+\beta_{1 Q}^{+} R Q_{t-1}^{1 / 2}\right) R V_{t-1}^{+}+\left(\beta_{1}^{-}+\beta_{1 Q}^{-} R Q_{t-1}^{1 / 2}\right) R V_{t-1}^{-} \\ & +\beta_{2} R V_{t-1 \mid t-5}+\beta_{3} R V_{t-1 \mid t-22}+u_{t}. \end{aligned} \end{equation}$

(18)

Table 9 compares out-of-sample forecast results from each of the alternative Q-models (HARQ-J, CHARQ, and SHARQ), to their non-Q adjusted baseline specification. We also include the HARQ model. For both currencies, the enhancements seen in the HARQ-J and CHARQ models align with those observed in the basic HARQ model. This is in contrast to the SHARQ model, which is outperformed by SHAR. Bollerslev et al. (2016) report similar results.

Table 9. Out-of-sample forecast losses for alternative Q-models.

EURUSD	HARQ	HARQ-J	CHARQ	SHARQ
MSE-RW	0.9759	0.9693	0.9749	1.0613
MSE-IW	0.9742	0.9563	0.9567	1.0315
QLIKE-RW	0.9767	0.9845	0.9750	1.1473
QLIKE-IW	0.9952	0.9960	0.9893	0.9987
USDJPY	HARQ	HARQ-J	CHARQ	SHARQ
MSE-RW	0.8885	0.8916	0.8914	1.0953
MSE-IW	0.9446	0.9322	0.9389	0.8965
QLIKE-RW	0.8824	0.8471	0.9040	1.3887
QLIKE-IW	0.8949	0.8942	0.9178	0.8974
Note: Model performance, expressed as model loss normalized by the loss of the relevant baseline models without the Q-adjustment terms. Each row reflects a combination of estimation window and loss function. Ratio for the best performing model on each row in bold.

| Show Table

DownLoad: CSV

3.3.3. Subsample analysis

Recent history contains two independent events that separately have induced turbulence in the global macroeconomy and financial markets. One is the outbreak of COVID-19 in March 2020; another is the Russian invasion of Ukraine in the second half of 2022, as illustrated in Figure 1.

Figure 1. EURUSD realized variance.

DownLoad: Full-Size Img PowerPoint

To analyze this period of extreme market conditions in isolation, we perform a sub-sample analysis covering 2020–2022. Table 10 contains out-of-sample results for day-ahead volatility forecasts. Reassuringly, the overall results remain intact, in that the HARQ-F model is the best performing model also when this extreme period is considered in isolation.

Table 10. Day ahead out-of-sample forecast losses, 2020–2022 subsample.

EURUSD	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.2522	1.0000	0.9781	0.9745	1.0041	1.0425	0.9517	0.9304
MSE-IW	1.2068	1.0000	0.9813	0.9764	0.9979	1.0976	0.9806	0.9677
QLIKE-RW	1.3216	1.0000	1.0169	0.9829	1.0093	1.1370	0.9446	0.9065
QLIKE-IW	1.5585	1.0000	1.0085	1.0119	1.0059	1.2338	0.9725	0.9701
USDJPY	AR	HAR	HAR-J	CHAR	SHAR	ARQ	HARQ	HARQ-F
MSE-RW	1.0930	1.0000	1.0555	0.9909	0.9822	0.9895	0.9564	0.9348
MSE-IW	1.1099	1.0000	0.9958	0.9850	1.0112	1.0523	1.0071	0.9827
QLIKE-RW	1.3404	1.0000	1.2635	1.0136	0.9845	0.9509	0.8611	0.8677
QLIKE-IW	1.4766	1.0000	0.9939	0.9808	1.0108	1.0231	0.8453	0.7868
Note: Model performance, expressed as model loss normalized by the loss of the HAR model. Each row reflects a combination of estimation window and loss function. Ratio for the best performing model on each row in bold.

| Show Table

DownLoad: CSV

4. Conclusions

This study uses updated tick-level data from two major currency pairs, EURUSD and USDJPY, covering January 2010 to December 2022, to investigate the relevance of realized quarticity for out-of-sample volatility forecasts. We find that realized quarticity effectively captures noise caused by measurement errors, as evidenced by increased precision in daily, weekly, and monthly volatility estimates from models augmented with realized quarticity as an additionally explanatory variable. These results are robust across estimation windows, evaluation metrics, and model specifications. As such, the results conform to comparable studies from other markets, predominantly on equity indices and single stocks. This paper also complements the relatively scarce body of literature on foreign exchange markets in this context.

A myriad of volatility models based on the HAR framework have been proposed. Still, simple linear HAR specifications have proven remarkably difficult to beat, as shown by Audrino et al. (2024) and Branco et al. (2024). In a recent survey, Gunnarsson et al. (2024) report promising results for machine learning models and volatility forecasting across asset classes. The FX implied volatility surface contains a rich set of relevant predictive information across forecasting horizons and quantiles (de Lange et al., 2022). Thus, combining implied volatilities and high-frequency data using machine learning models, along the lines of Blom et al. (2023), appears as an interesting avenue for future research.

Rarely, one single model dominates others in terms of statistical and economic criteria. To this end, investigating ensemble models where high-frequency models are combined with other volatility model classes, such as time series models and stochastic volatility models-possibly including jump-processes, should be of interest. The recently developed rough-path volatility models based on fractional Brownian motion (Salmon and SenGuptz, 2021; Bayer et al., 2023) appear particularly relevant in this context.

Use of AI tools declaration

The authors declare they have not used Artificial Intelligence (AI) tools in the creation of this article.

Author contributions

M.R.: Conceptualization, Methodology, Software, Formal analysis, Writing - Original Draft, Writing - Review & Editing.

M.H.: Data Curation, Writing - Original Draft, Writing - Review & Editing.

Acknowledgments

We would like to thank Andrew Patton for making the Matlab code from Bollerslev et al. (2016) available at https://public.econ.duke.edu/ap172/. Furthermore, we are grateful for insightful comments from the Editor and two anonymous reviewers, which helped us improve the paper.

Conflict of interest

The authors declare no conflicts of interest.

References

[1]	C. Depcik, T. Cassady, B. Collicott, S. P. Burugupally, J. Hobeck, Comparison of lithium-ion ion batteries, hydrogen fueled combustion Engines, and a hydrogen fuel cell in powering a small Unmanned Aerial Vehicle, Energy Convers. Manage., 207 (2020), 112514. https://doi.org/10.1016/j.enconman.2020.112514 doi: 10.1016/j.enconman.2020.112514
[2]	M. Chen, G. Rincon-Mora, Accurate electrical battery model capable of predicting runtime and Ⅰ–Ⅴ performance, IEEE Trans. Power Syst., 21 (2006), 504–511. https://doi.org/10.1109/TEC.2006.874229 doi: 10.1109/TEC.2006.874229
[3]	J. B. Goodenough, K. S. Park, The li-ion rechargeable battery: A perspective, J. Am. Chem. Soc., 135 (2013), 1167–1176. https://doi.org/10.1021/ja3091438 doi: 10.1021/ja3091438
[4]	Z. Liu, B. He, Z. Zhang, W. Deng, D. Dong, S. Xia, et al., Lithium/graphene composite anode with 3D structural LiF protection layer for high-performance lithium metal batteries, ACS Appl. Mater. Interfaces., 14 (2022), 2871–2880. https://doi.org/10.1021/acsami.1c21263 doi: 10.1021/acsami.1c21263
[5]	A. Attanayaka, J. Karunadasa, K. Hemapala, Estimation of state of charge for lithium-ion batteries-A review, AIMS Energy, 7 (2019), 186–210. https://doi.org/10.3934/energy.2019.2.186 doi: 10.3934/energy.2019.2.186
[6]	A. Basia, Z. Simeu-Abazi, E. Gascard, P. Zwolinski, Review on state of health estimation methodologies for lithium-ion batteries in the context of circular economy, CIRP J. Manuf. Sci. Technol., 32 (2021), 517–528. https://doi.org/10.1016/j.cirpj.2021.02.004 doi: 10.1016/j.cirpj.2021.02.004
[7]	C. Julien, A. Mauger, A. Abdel-Ghany, A. Hashem, K. Zaghib, Smart materials for energy storage in Li-ion batteries, AIMS Mater. Sci., 3 (2016), 137–148. https://doi.org/10.3934/matersci.2016.1.137 doi: 10.3934/matersci.2016.1.137
[8]	M. Ge, Y. Liu, X. Jiang, J. Liu, A review on state of health estimations and remaining useful life prognostics of lithium-ion batteries, Measurement, 174 (2021), 109057. https://doi.org/10.1016/j.measurement.2021.109057 doi: 10.1016/j.measurement.2021.109057
[9]	C. Hu, B. Youn, P. Wang, J. K. Yoon, Ensemble of data-driven prognostic algorithms for robust prediction of remaining useful life, Reliab. Eng. Syst. Saf., 103 (2012), 120–135. https://doi.org/10.1016/j.ress.2012.03.008 doi: 10.1016/j.ress.2012.03.008
[10]	S. Jarid, M. Das, An electro-thermal model based fast optimal charging strategy for lithium-ion batteries, AIMS Energy, 9 (2021), 915–933. https://doi.org/10.3934/energy.2021043 doi: 10.3934/energy.2021043
[11]	G. Ma, Y. Zhang, C. Cheng, B. Zhou, P. Hu, Y. Yuan, Remaining useful life prediction of lithium-ion batteries based on false nearest neighbors and a hybrid neural network, Appl. Energy, 253 (2019), 113626. https://doi.org/10.1016/j.apenergy.2019.113626 doi: 10.1016/j.apenergy.2019.113626
[12]	L. Wu, X. Fu, Y. Guan, Review of the remaining useful life prognostics of vehicle lithium-ion batteries using data-driven methodologies, Appl. Sci., 6 (2016), 166. https://doi.org/10.3390/app6060166 doi: 10.3390/app6060166
[13]	A. Nuhic, T. Terzimehic, T. Soczka-Guth, M. Buchholz, K. Dietmayer, Health diagnosis and remaining useful life prognostics of lithium-ion batteries using data-driven methods, J. Power Sources, 239 (2013), 680–688. https://doi.org/10.1016/j.jpowsour.2012.11.146 doi: 10.1016/j.jpowsour.2012.11.146
[14]	S. Wang, S. Jin, D. Bai, Y. Fan, H. Shi, C. Fernandez, A critical review of improved deep learning methods for the remaining useful life prediction of lithium-ion batteries, Energy Rep., 7 (2021), 5562–5574. https://doi.org/10.1016/j.egyr.2021.08.182 doi: 10.1016/j.egyr.2021.08.182
[15]	N. Khare, P. Singh, J. K. Vassiliou, A novel magnetic field prob-ing technique for determining state of health of sealed lead-acid batteries, J. Power Sources, 218 (2012), 462–473. https://doi.org/10.1016/j.jpowsour.2012.06.085 doi: 10.1016/j.jpowsour.2012.06.085
[16]	A. Mevawalla, Y. Shabeer, M. K. Tran, S. Panchal, M. Fowler, R. Fraser, Thermal modelling utilizing multiple experimentally measurable parameters, Batteries, 8 (2022), 147. https://doi.org/10.3390/batteries8100147 doi: 10.3390/batteries8100147
[17]	Y. Wang, D. Dan, Y. Zhang, Y. Qian, S. Panchal, M. Fowler, et al., A novel heat dissipation structure based on flat heat pipe for battery thermal management system, Int. J. Energy Res., 46 (2022), 15961–15980. https://doi.org/10.1002/er.8294 doi: 10.1002/er.8294
[18]	Y. Xie, W. Li, X. Hu, M. K. Tran, S. Panchal, M. Fowler, et al., Co-estimation of SOC and three-dimensional SOT for lithium-ion batteries based on distributed spatial-temporal online correction, IEEE Trans. Ind. Electron., 2022 (2022), 1–10. https://doi.org/10.1109/TIE.2022.3199905 doi: 10.1109/TIE.2022.3199905
[19]	Y. Xing, N. Williard, K. L. Tsui, M. Pecht, A comparative review of prognostics-based reliability methods for Lithium batteries, in 2011 Prognostics and System Health Managment Confernece, 2011. https://doi.org/10.1109/PHM.2011.5939585
[20]	D. Wang, F. Yang, K. L. Tsui, Q. Zhou, B. S. Bae, Remaining useful life prediction of lithium-ion batteries based on spherical cubature particle filter, IEEE Trans. Instrum. Meas., 65 (2016), 1282–1291. https://doi.org/10.1109/TIM.2016.2534258 doi: 10.1109/TIM.2016.2534258
[21]	M. K. Tran, A. DaCosta, A. Mevawalla, S. Panchal, M. Fowler, Comparative study of equivalent circuit models performance in four common lithium-ion batteries: LFP, NMC, LMO, NCA, Batteries, 7 (2021), 51. https://doi.org/10.3390/batteries7030051 doi: 10.3390/batteries7030051
[22]	Z. Lyu, R. Gao, L. Chen, Li-ion battery state of health estimation and remaining useful life prediction through a model-data-fusion method, IEEE Trans. Power Electron., 36 (2021), 6228–6240. https://doi.org/10.1109/TPEL.2020.3033297 doi: 10.1109/TPEL.2020.3033297
[23]	S. Wang, P. Ren, P. Takyi-Aninakwa, S. Jin, C. Fernandez, A critical review of improved deep convolutional neural network for multi-timescale state prediction of lithium-ion batteries, Energies, 15 (2022), 5053. https://doi.org/10.3390/en15145053 doi: 10.3390/en15145053
[24]	S. Jin, X. Sui, X. Huang, S. Wang, R. Teodorescu, D. I. Stroe, Overview of machine learning methods for lithium-ion battery remaining useful lifetime prediction, Electronics, 10 (2021), 3126. https://doi.org/10.3390/electronics10243126 doi: 10.3390/electronics10243126
[25]	P. Khumprom, N. Yodo, A data-driven predictive prognostic model for lithium-ion batteries based on a deep learning algorithm, Energies, 12 (2019), 660. https://doi.org/10.3390/en12040660 doi: 10.3390/en12040660
[26]	L. Cai, J. Meng, D. I. Stroe, J. Peng, R. Teodorescu, Multi-objective optimization of data-driven model for lithium-ion battery SOH estimation with short-term feature, IEEE Trans. Power Electron., 35 (2020), 11855–11864. https://doi.org/10.1109/TPEL.2020.2987383 doi: 10.1109/TPEL.2020.2987383
[27]	T. Qin, S. Zeng, J. Guo, Robust prognostics for state of health estimation of lithium-ion batteries based on an improved PSO-SVR model, Microelectron. Reliab., 55 (2015), 1280–1284. https://doi.org/10.1016/j.microrel.2015.06.133 doi: 10.1016/j.microrel.2015.06.133
[28]	Y. Cai, Y. Lin, Z. Deng, X. Zhao, D. Hao, Prediction of lithium-ion battery remaining useful life based on hybrid data-driven method with optimized parameter, in 2017 2nd International Conference on Power and Renewable Energy (ICPRE), 2017. https://doi.org/10.1109/ICPRE.2017.8390489
[29]	B. Gou, Y. Xu, X. Feng, State-of-health estimation and remaining useful life prediction for lithium-ion battery using a hybrid data-driven method, IEEE Trans. Veh. Technol., 69 (2020), 10854–10867. https://doi.org/10.1109/TVT.2020.3014932 doi: 10.1109/TVT.2020.3014932
[30]	G. Ma, Y. Zhang, C. Cheng, B. Zhou, Y. Yuan, Remaining useful life prediction of lithium-ion batteries based on false nearest neighbors and a hybrid neural network, Appl. Energy, 253 (2019), 113626. https://doi.org/10.1016/j.apenergy.2019.113626 doi: 10.1016/j.apenergy.2019.113626
[31]	Y. Zhang, R. Xiong, H. He, M. Pecht, Long short-term memory recurrent neural network for remaining useful life prediction of lithium-ion batteries, IEEE Trans. Veh. Technol., 67 (2018), 5695–5705. https://doi.org/10.1109/TVT.2018.2805189 doi: 10.1109/TVT.2018.2805189
[32]	S. Yalçın, S. Panchal, M. S. Herdem, A CNN-ABC model for estimation and optimization of heat generation rate and voltage distributions of lithium-ion batteries for electric vehicles, Int. J. Heat Mass. Tran., 199 (2022), 123486. https://doi.org/10.1016/j.ijheatmasstransfer.2022.123486 doi: 10.1016/j.ijheatmasstransfer.2022.123486
[33]	F. Wang, Z. Zhao, J. Ren, Z. Zhai, S. Wang, X. Chen, A transferable lithium-ion battery remaining useful life prediction method from cycle-consistency of degradation trend, J. Power Sources, 521 (2022), 230975. https://doi.org/10.1016/j.jpowsour.2022.230975 doi: 10.1016/j.jpowsour.2022.230975
[34]	S. Wang, P. Takyi-Aninakwa, S. Jin, C. Yu, C. Fernandez, D. I. Stroe, An improved feedforward-long short-term memory modeling method for the whole-life-cycle state of charge prediction of lithium-ion batteries considering current-voltage-temperature variation, Energy, 254 (2022), 124224. https://doi.org/10.1016/j.energy.2022.124224 doi: 10.1016/j.energy.2022.124224
[35]	M. Xia, X. Zheng, M. Imran, M. Shoaib, Data-driven prognosis method using hybrid deep recurrent neural network, Appl. Soft Comput., 93 (2020), 106351. https://doi.org/10.1016/j.asoc.2020.106351 doi: 10.1016/j.asoc.2020.106351
[36]	A. Kara, A data-driven approach based on deep neural networks for lithium-ion battery prognostics, Neural Comput. Appl., 33 (2021), 13525–13538. https://doi.org/10.1007/s00521-021-05976-x doi: 10.1007/s00521-021-05976-x
[37]	C. Wang, N. Lu, S. Wang, Y. Cheng, B. Jiang, Dynamic long short-term memory neural-network-based indirect remaining-useful-life prognosis for satellite lithium-ion battery, Appl. Sci., 8 (2018), 2078. https://doi.org/10.3390/app8112078 doi: 10.3390/app8112078
[38]	P. Li, Z. Zhang, Q. Xiong, B. Ding, S. Li, State-of-health estimation and remaining useful life prediction for the lithium-ion battery based on a variant long short term memory neural network, J. Power Sources, 459 (2020), 228069. https://doi.org/10.1016/j.jpowsour.2020.228069 doi: 10.1016/j.jpowsour.2020.228069
[39]	M. Geraldi, E. Ghisi, Data-driven framework towards realistic bottom-up energy benchmarking using an Artificial Neural Network, Appl. Energy, 306 (2022), 117960. https://doi.org/10.1016/j.apenergy.2021.117960 doi: 10.1016/j.apenergy.2021.117960
[40]	R. Lei, Z. Li, H. Sheng, S. Zhao, W. Hao, Z. Lin, Remaining useful life prediction for lithium-ion battery: A deep learning approach, IEEE Access, 6 (2018), 50587–50598. https://doi.org/10.1109/ACCESS.2018.2858856 doi: 10.1109/ACCESS.2018.2858856
[41]	N. Harting, R. Schenkendorf, N. Wolff, U. Krewer, State-of-health identification of lithium-ion batteries based on nonlinear frequency response analysis: First steps with machine learning, Appl. Sci., 8 (2018), 821. https://doi.org/10.3390/app8050821 doi: 10.3390/app8050821
[42]	B. Zraibi, C. Okar, H. Chaoui, M. Mansouri, Remaining useful life assessment for lithium-ion batteries using CNN-LSTM-DNN hybrid method, IEEE Trans. Veh. Technol., 70 (2021), 4252–4261. https://doi.org/10.1109/TVT.2021.3071622 doi: 10.1109/TVT.2021.3071622
[43]	Y. Anagun, S. Isik, E. Seke, SRLibrary: Comparing different loss functions for super-resolution over various convolutional architectures, J. Visual Commun. Image Represent., 61 (2019), 178–187. https://doi.org/10.1016/j.jvcir.2019.03.027 doi: 10.1016/j.jvcir.2019.03.027
[44]	Y. Zhou, M. Huang, Lithium-ion batteries remaining useful life prediction based on a mixture of empirical mode decomposition and ARIMA model, Microelectron. Reliab., 65 (2016), 265–273. https://doi.org/10.1016/j.microrel.2016.07.151 doi: 10.1016/j.microrel.2016.07.151
[45]	R. Sekhar, P. Shah, S. Panchal, M. Fowler, R. Fraser, Distance to empty soft sensor for ford escape electric vehicle, Results Control Optim., 9 (2022), 100168, https://doi.org/10.1016/j.rico.2022.100168 doi: 10.1016/j.rico.2022.100168
[46]	M. Li, J. Lu, Z. Chen, K. Amine, 30 years of lithium-ion batteries, Adv. Mater., 30 (2018), 1800561. https://doi.org/10.1002/adma.201800561 doi: 10.1002/adma.201800561
[47]	M. M. Kabir, D. E. Demirocak, Degradation mechanisms in Li-ion batteries: a state-of-the-art review, Int. J. Energy Res., 41 (2017), 1963–1986. https://doi.org/10.1002/er.3762 doi: 10.1002/er.3762
[48]	J. Vetter, P. Novák, M. R. Wagner, C. Veit, K. C. Möller, J. O. Besenhard, et al., Ageing mechanisms in lithium-ion batterie, J. Power Sources, 147 (2005), 269–281. https://doi.org/10.1016/j.jpowsour.2005.01.006 doi: 10.1016/j.jpowsour.2005.01.006
[49]	B. Saha, K. Goebel, Battery data set, in NASA Ames Prognostics Data Repository, 2007. Available from: http://ti.arc.nasa.gov/project/prognostic-data-repository.
[50]	L. Alzubaidi, J. Zhang, A. J. Humaidi, A. Al-Dujaili, Y. Duan, O. Al-Shamma, et al., Review of deep learning: Concepts, CNN architectures, challenges, applications, future directions, J. Big Data, 8 (2021), 1–74. https://doi.org/10.1186/s40537-021-00444-8 doi: 10.1186/s40537-021-00444-8
[51]	D. Yao, B. Li, H. Liu, J. Yang, L. Jia, Remaining useful life prediction of roller bearings based on improved 1D-CNN and simple recurrent unit, Measurement, 175 (2021), 109166. https://doi.org/10.1016/j.measurement.2021.109166 doi: 10.1016/j.measurement.2021.109166
[52]	A. Sherstinsky, Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) network, Physica D, 404 (2020), 132306. https://doi.org/10.1016/j.physd.2019.132306 doi: 10.1016/j.physd.2019.132306
[53]	Z. Shi, A. Chehade, A dual-LSTM framework combining change point detection and remaining useful life prediction, Reliab. Eng. Syst. Saf., 205 (2021), 107257. https://doi.org/10.1016/j.ress.2020.107257 doi: 10.1016/j.ress.2020.107257
[54]	Y. Choi, S. Ryu, K. Park, H. Kim, Machine learning-based lithium-ion battery capacity estimation exploiting multi-channel charging profiles, IEEE Access, 7 (2019), 75143–75152. https://doi.org/10.1109/ACCESS.2019.2920932 doi: 10.1109/ACCESS.2019.2920932
[55]	X. Hu, J. Jiang, D. Cao, B. Egardt, Battery health prognosis for electric vehicles using sample entropy and sparse bayesian predictive modeling, IEEE Trans. Ind. Electron., 63 (2016), 2645–2656. https://doi.org/10.1109/TIE.2015.2461523 doi: 10.1109/TIE.2015.2461523

Reader Comments

Your name:*

Email:*
© 2023 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)