
Within the domain of cardiovascular diseases, arrhythmia is one of the leading anomalies causing sudden deaths. These anomalies, including arrhythmia, are detectable through the electrocardiogram, a pivotal component in the analysis of heart diseases. However, conventional methods like electrocardiography encounter challenges such as subjective analysis and limited monitoring duration. In this work, a novel hybrid model, AttBiLFNet, was proposed for precise arrhythmia detection in ECG signals, including imbalanced class distributions. AttBiLFNet integrates a Bidirectional Long Short-Term Memory (BiLSTM) network with a convolutional neural network (CNN) and incorporates an attention mechanism using the focal loss function. This architecture is capable of autonomously extracting features by harnessing BiLSTM's bidirectional information flow, which proves advantageous in capturing long-range dependencies. The attention mechanism enhances the model's focus on pertinent segments of the input sequence, which is particularly beneficial in class imbalance classification scenarios where minority class samples tend to be overshadowed. The focal loss function effectively addresses the impact of class imbalance, thereby improving overall classification performance. The proposed AttBiLFNet model achieved 99.55% accuracy and 98.52% precision. Moreover, performance metrics such as MF1, K score, and sensitivity were calculated, and the model was compared with various methods in the literature. Empirical evidence showed that AttBiLFNet outperformed other methods in terms of both accuracy and computational efficiency. The introduced model serves as a reliable tool for the timely identification of arrhythmias.
Citation: Enes Efe, Emrehan Yavsan. AttBiLFNet: A novel hybrid network for accurate and efficient arrhythmia detection in imbalanced ECG signals[J]. Mathematical Biosciences and Engineering, 2024, 21(4): 5863-5880. doi: 10.3934/mbe.2024259
[1] | Xinyu Shao, Zhen Liu, Baoping Jiang . Sliding-mode controller synthesis of robotic manipulator based on a new modified reaching law. Mathematical Biosciences and Engineering, 2022, 19(6): 6362-6378. doi: 10.3934/mbe.2022298 |
[2] | Xiangfei Meng, Guichen Zhang, Qiang Zhang . Robust adaptive neural network integrated fault-tolerant control for underactuated surface vessels with finite-time convergence and event-triggered inputs. Mathematical Biosciences and Engineering, 2023, 20(2): 2131-2156. doi: 10.3934/mbe.2023099 |
[3] | Xiaoqiang Dai, Hewei Xu, Hongchao Ma, Jianjun Ding, Qiang Lai . Dual closed loop AUV trajectory tracking control based on finite time and state observer. Mathematical Biosciences and Engineering, 2022, 19(11): 11086-11113. doi: 10.3934/mbe.2022517 |
[4] | Yi Zhang, Yue Song, Song Yang . T-S fuzzy observer-based adaptive tracking control for biological system with stage structure. Mathematical Biosciences and Engineering, 2022, 19(10): 9709-9729. doi: 10.3934/mbe.2022451 |
[5] | Chengxi Wu, Yuewei Dai, Liang Shan, Zhiyu Zhu, Zhengtian Wu . Data-driven trajectory tracking control for autonomous underwater vehicle based on iterative extended state observer. Mathematical Biosciences and Engineering, 2022, 19(3): 3036-3055. doi: 10.3934/mbe.2022140 |
[6] | Chao Wang, Cheng Zhang, Dan He, Jianliang Xiao, Liyan Liu . Observer-based finite-time adaptive fuzzy back-stepping control for MIMO coupled nonlinear systems. Mathematical Biosciences and Engineering, 2022, 19(10): 10637-10655. doi: 10.3934/mbe.2022497 |
[7] | Xingjia Li, Jinan Gu, Zedong Huang, Chen Ji, Shixi Tang . Hierarchical multiloop MPC scheme for robot manipulators with nonlinear disturbance observer. Mathematical Biosciences and Engineering, 2022, 19(12): 12601-12616. doi: 10.3934/mbe.2022588 |
[8] | Vladimir Djordjevic, Hongfeng Tao, Xiaona Song, Shuping He, Weinan Gao, Vladimir Stojanovic . Data-driven control of hydraulic servo actuator: An event-triggered adaptive dynamic programming approach. Mathematical Biosciences and Engineering, 2023, 20(5): 8561-8582. doi: 10.3934/mbe.2023376 |
[9] | Na Zhang, Jianwei Xia, Tianjiao Liu, Chengyuan Yan, Xiao Wang . Dynamic event-triggered adaptive finite-time consensus control for multi-agent systems with time-varying actuator faults. Mathematical Biosciences and Engineering, 2023, 20(5): 7761-7783. doi: 10.3934/mbe.2023335 |
[10] | Chaoyue Wang, Zhiyao Ma, Shaocheng Tong . Adaptive fuzzy output-feedback event-triggered control for fractional-order nonlinear system. Mathematical Biosciences and Engineering, 2022, 19(12): 12334-12352. doi: 10.3934/mbe.2022575 |
Within the domain of cardiovascular diseases, arrhythmia is one of the leading anomalies causing sudden deaths. These anomalies, including arrhythmia, are detectable through the electrocardiogram, a pivotal component in the analysis of heart diseases. However, conventional methods like electrocardiography encounter challenges such as subjective analysis and limited monitoring duration. In this work, a novel hybrid model, AttBiLFNet, was proposed for precise arrhythmia detection in ECG signals, including imbalanced class distributions. AttBiLFNet integrates a Bidirectional Long Short-Term Memory (BiLSTM) network with a convolutional neural network (CNN) and incorporates an attention mechanism using the focal loss function. This architecture is capable of autonomously extracting features by harnessing BiLSTM's bidirectional information flow, which proves advantageous in capturing long-range dependencies. The attention mechanism enhances the model's focus on pertinent segments of the input sequence, which is particularly beneficial in class imbalance classification scenarios where minority class samples tend to be overshadowed. The focal loss function effectively addresses the impact of class imbalance, thereby improving overall classification performance. The proposed AttBiLFNet model achieved 99.55% accuracy and 98.52% precision. Moreover, performance metrics such as MF1, K score, and sensitivity were calculated, and the model was compared with various methods in the literature. Empirical evidence showed that AttBiLFNet outperformed other methods in terms of both accuracy and computational efficiency. The introduced model serves as a reliable tool for the timely identification of arrhythmias.
Ap | Ram area |
Ct | Internal leakage coefficient |
P1, P2, | Oil pressures inside the chambers of the actuator, respectively |
PL | Load pressure |
Pr | Return pressure |
Ps | Supply pressure |
Q1, Q2 | Oil flow in both chambers of the actuator |
V01, V02 | Initial volume of oil inlet and outlet chamber, respectively |
V1, V2 | Control volume of oil inlet and outlet chamber, respectively |
fv | Damping coefficient |
g1, g2 | Pressure difference of inlet and outlet |
k | Spring stiffness |
k1, k2, k3, kr | Positive gains |
kt | Total gain of the servo valve |
mp | Mass of load |
q1, q2 | Model errors because of internal leakage |
u | Control voltage |
xp | Position of load |
β | Gain |
βe1, βe2 | Oil effective bulk modulus of the actuator, respectively |
ω | Bandwidth of the extended state observer |
ARISE | Adaptive robust integral of the sign of the error |
ESO | Extended state observer |
RISE | Robust integral of the sign of the error |
Electric hydraulic servo system (EHSS) is widely used in industry applications due to its high power ratio and fast response [1,2,3,4,5,6]. With the rapid development of technology, more severe control requirements are put forward for EHSS, and high-performance control becomes eagerly needed to address its behavioral nonlinearity and uncertainties, such as flow nonlinearity, pressure dynamic nonlinearity, uncertain parameters, unmolded nonlinearity, and external disturbances [7,8,9,10].
Nowadays, many researchers focus on advanced control strategies such as robust adaptive control [11,12], sliding mode control [13,14], back-stepping control [15,16], robust integral of the sign of the error (RISE) [17,18] for EHSS. Yue and Yao [17] proposed an adaptive robust integral of the sign of the error control (ARISE), which can adjust the robust gain online through adaptive method to solve the potential high gain feedback of symbolic function. Yao et al. [19] designed a novel ARSE to address noise pollution in the acquisition of acceleration signal, which can compensate the error of friction model and other bounded disturbances [20]. The above literature shows that RISE/ARISE control greatly improves the trajectory tracking accuracy of EHSS. However, those controllers regard the parameter adaptive error, unmolded error, and external disturbance as lumped disturbance, which severely limits the control accuracy. Control strategy based on disturbance observer can compensate the influence of disturbance and uncurtains effectively, which has been used in the field of control theory and engineering [21,22,23,24,25,26]. Especially, the extended state observer (ESO), which is the core of ADRC and has been widely used in disturbance estimation and suppression [27,28,29,30,31,32].
In this paper, the sign function is replaced with a modified arctangent function to smooth the nonlinearity of sign function. Then, the state and external disturbance can be estimated by ESO, respectively. Finally, the residual observation error is compensated to further enhance the tracking accuracy by ARISE. The Lyapunov theory proves the EHSS can achieve asymptotic s. Simulation results show the proposed controller has a better performance in maxi-mum tracking error, average tracking error and standard deviation of the tracking error.
The typical working principle of EHSS is shown in Figure 1. The double rod symmetrical hydraulic cylinder is controlled by servo valve to drive the load.
In this paper, the force balance equation of EHSS can be given as
mp¨xp=PLAp−kxp−fv˙xp+f(xp,˙xp,t) | (1) |
where mp and xp represent mass and displacement of load respectively; PL is the pressure difference; Ap is the effective area of the piston; k is the spring stiffness; fv is the combined coefficient; f(xp,˙xp,t) indicates the lumped uncertain; Pressure dynamics of the two chambers are given by [15].
{˙P1=βe1/βe1V1V1(Q1−Ap˙xp−CtPL+q1(t))˙P2=βe2/βe2V2V2(Ap˙xp+CtPL−Q2−q2(t)) | (2) |
where
{Q1=ktg1(P1,xv)uQ2=ktg2(P2,xv)u | (3) |
{g1=√(Ps−Pt+(Ps−2P1+Pt)⋅s(u))/2g2=√(Ps−Pt−(Ps−2P2+Pt)⋅s(u))/2 | (4) |
The pressure difference dynamics between the two chambers is expressed as follows
˙PL=βekt(g1/g1V1+g2/g2V2V2V1+g2/g2V2V2)u−βeAp(1/1V1+1/1V2V2V1+1/1V2V2)˙xp−βeCt(1/1V1+1/1V2V2V1+1/1V2V2)PL+βe(q1/q1V1+q2/q2V2V2V1+q2/q2V2V2) | (5) |
where βe1, βe2 are the effective elastic modulus in two chamber and βe1 = βe2 = βe; V1 = V01 + Ap; V2 = V02 - Ap·xp represents the control volume of return chamber; V01 and V02 are the initial volumes of the two chambers respectively; Ct is the internal leakage coefficient of the cylinder; Q1 (t) and Q2 (t) are the oil flowrate of the two chamber of the cylinder respectively; q1(t) and q2(t) are model errors because of internal leakage in the two chambers; g1 is the pressure difference at the oil inlet and g2 is the pressure difference at the oil outlet; Ps is the supplied pressure; Pr is the return pressure, kt is the total gain of the servo valve; u is the control voltage and s(u) is expressed as
s(u)={1u>00u=0−1u<0 | (6) |
To make it smooth and differentiable, the sign function s(u) is replaced by Eq (7).
farctan(u)=2arctan(Ku)/π | (7) |
Thus, g1, g2 in Eq (4) can be rewritten as Eq (8).
{g′1=√(Ps−Pt+(Ps−2P1+Pt)⋅farctan(u))/2g′2=√(Ps−Pt−(Ps−2P2+Pt)⋅farctan(u))/2 | (8) |
Substituting Eqs (2)-(5) into Eq (1), and thus
mp⃛xp=Apβukl(g1/V1+g2/V2)u−A2pβu(1/V1+1/V2)ˉxp−ApβeCt(1/V1+1/V2)PL+Apβu(q1/V1+q2/V2)−k˙xp−fv¨xp+˙f | (9) |
Rewritten Eq (9), and thus,
mp⃛xp=Apβukt(g′1/V1+g2/V2)u−A2pβe(1/V1+1/V2)ˉxp−ApβeCt(1/V1+1/V2)FL+Apβe(q1/V1+q2/V2)−k˙xp−fv¨xp+˙f+f′ | (10) |
where f′ is the approximation error caused by using the continuously differentiable function farctan(u).
According to Eq (1)
PL=mpAp¨xp+kApxp+fvAp˙xp−fAp | (11) |
Substituting Eq (11) into Eq (10), thus
mp⃛xp=Apβekt(g′1/V1+g′2/V2)u−βeCt(1/V1+1/V2)kxp−(A2pβe(1/V1+1/V2)+fvβeCt(1/V1+1/V2)+k)˙xp−(mpβeCt(1/V1+1/V2)+fv)¨xp+βeCt(1/V1+1/V2)f+˙f+f′+Apβe(q1/V1+q2/V2) | (12) |
Define state variables as x=[x1,x2,x3]T=[xp,˙xp,¨xp]T and output variables as y=x1=xp, so the state space model of EHSS can be expressed as
{˙x1=x2˙x2=x3˙x3=θ1u−θ2x1−θ3x2−θ4x3+Δ | (13) |
where θ1=Apβekt(g′1V1+g′1V2)/mp,θ2=βeCt(1V1+1V2)k/mp, θ3=(A2pβe(1V1+1V2)+fvβeCt(1V1+1V2)+k)/mp, θ4=(mpβeCt(1V1+1V2)+fv)/mp,Δ=(βeCt(1V1+1V2)f+˙f+f′+Apβe(q1V1+q1V2))/mp.
In practice, the parameters m, k, βe and Ct may not be known accurately, so it is necessary to consider the uncertainties of these parameters. Define vector as θ=[θ1,θ2,θ3,θ4] and improve the tracking performance of the system through the adaptive method.
The purpose of the system controller is to design a bounded control input u so that y = x1 can track the desired trajectory yd(t) = x1d(t). Therefore, the following assumptions should be given.
Assumption 1: The desired trajectory x1d is five times differentiable and each is bounded. In practice, the load pressure of hydraulic cylinder meets 0 < PL < Ps.
Assumption 2: The range of parametric uncertainties is
θ∈Ωθ | (14) |
ΩθΔ{θ:θmin | (15) |
where \theta_{\min } = \left[\theta_{1 \min }, \cdots, \theta_{4 \min }\right]^{T} , \theta_{\max } = \left[\theta_{1 \max }, \cdots, \theta_{4 \max }\right]^{T} are known.
Assumption 3: The time-varying perturbation Δ (t) of Eq (10) is smooth enough so that
\mid \dot{\Delta}(t)\mid \leq \delta_{1} \& \mid\ddot{\Delta}(t)\mid \leq \delta_{2} | (16) |
where δ1, δ2 are known positive constants.
ESO can estimate the uncertainty disturbance comprehensively, so we use ESO to estimate the lumped disturbance and compensate it feed forward to achieve better tracking accuracy.
Define x4 = Δ(t), {\dot x_4} = \delta \left(t \right) and the expanded state space can be written as
\left\{ \begin{array}{l} {{\dot x}_1} = {x_2} \hfill \\ {{\dot x}_2} = {x_3} \hfill \\ {{\dot x}_3} = {\theta _1}u - {\theta _2}{x_1} - {\theta _3}{x_2} - {\theta _4}{x_3} + {x_4} \hfill \\ {{\dot x}_4} = \delta \hfill \\ \end{array} \right. | (17) |
Defining \hat{x} = x-\hat{x} as the estimation error of x, where \hat{x} represents the estimated value of x.
According to the expanded state space model, the ESO is designed as
\left\{\begin{array}{l} \dot{\hat{x}}_1 = \hat{x}_2+4 \omega_0\left(x_1-\hat{x}_1\right) \\ \dot{\hat{x}}_2 = \hat{x}_3+6 \omega_0^2\left(x_1-\hat{x}_1\right) \\ \dot{\hat{x}}_3 = \hat{\theta}_1 u-\hat{\theta}_2 x_1-\hat{\theta}_3 x_2-\hat{\theta}_3 x_3+\hat{x}_4+4 \omega_0^3\left(x_1-\hat{x}_1\right) \\ \dot{\hat{x}}_4 = \omega_0^4\left(x_1-\hat{x}_1\right) \end{array}\right. | (18) |
where ω0 is the bandwidth of the extended state observer, \hat{\theta}_{i} is the estimated value of the unknown parameters θi, I = 1, 2, 3, 4.
The dynamic equation of observation error can be obtained by subtracting Eq (17) an Eq (18):
\left\{\begin{array}{l} \dot{\tilde{x}}_1 = \tilde{x}_2-4 \varpi_0 \tilde{x}_1 \\ \dot{\tilde{x}}_2 = \tilde{x}_3-6 \omega_0^2 \tilde{x}_1 \\ \dot{\tilde{x}}_3 = \tilde{x}_4+\tilde{\boldsymbol{\theta}}^{\mathrm{T}} \boldsymbol{\varPhi}_1-4 \omega_0^3 \tilde{x}_1 \\ \dot{\tilde{x}}_4 = \delta-\omega_0^4 \tilde{x}_1 \end{array}\right. | (19) |
where \boldsymbol{\tilde{\theta}} = \left[\tilde{\theta}_{1}, \tilde{\theta}_{2}, \tilde{\theta}_{3}, \tilde{\theta}_{4}\right]^{\mathrm{T}} , \boldsymbol{ \varPhi}_{1} = \left[u, x_{1}, x_{2}, x_{\xi}, x_{4}\right]^{T} .
Let
{\xi _i} = \frac{{{{\tilde x}_i}}}{{\omega _0^{i - 1}}}, i = 1, 2, 3, 4 | (20) |
Then Eq (17) can be written as
\left[\begin{array}{l} \dot{\xi}_1 \\ \dot{\xi}_2 \\ \dot{\xi}_3 \\ \dot{\xi}_4 \end{array}\right] = \varpi_0\underbrace{\left[\begin{array}{llll} -4 & 1 & 0 & 0 \\ -6 & 0 & 1 & 0 \\ -4 & 0 & 0 & 1 \\ -1 & 0 & 0 & 0 \end{array}\right]}_{\boldsymbol{A}}\left[\begin{array}{l} \xi_1 \\ \xi_2 \\ \xi_3 \\ \xi_4 \end{array}\right]+\underbrace{\left[\begin{array}{l} 0 \\ 0 \\ 1 \\ 0 \end{array}\right]}_{B_1} \frac{\tilde{\theta}^T \Phi}{\varpi_0^2}+\underbrace{\left[\begin{array}{l} 0 \\ 0 \\ 0 \\ 1 \end{array}\right]}_{B_2} \frac{\delta}{\varpi_0^3} | (21) |
\dot{\xi} = \omega_0 \boldsymbol{A} \xi+\boldsymbol{B}_1 \frac{\tilde{\boldsymbol{\theta}}^{\mathrm{T}} \boldsymbol{\varPhi}}{\omega_0^2}+\boldsymbol{B}_2 \frac{\delta}{\varpi_0^3} | (22) |
Since matrix A is a Hurwitz matrix, and there is a positive definite symmetric matrix P which satisfies the following equation
\boldsymbol{A}^{\mathrm{T}} \boldsymbol{P}+\boldsymbol{P} \boldsymbol{A} = -\boldsymbol{I} | (23) |
The symmetric positive definite matrix P is:
\boldsymbol{P} = \left[ {\begin{array}{*{20}{c}} {{{17} \mathord{\left/ {\vphantom {{17} 8}} \right. } 8}}&{ - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{ - {{11} \mathord{\left/ {\vphantom {{11} 8}} \right. } 8}}&{{1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}} \\ { - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{{{11} \mathord{\left/ {\vphantom {{11} 8}} \right. } 8}}&{ - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{{{ - 17} \mathord{\left/ {\vphantom {{ - 17} 8}} \right. } 8}} \\ { - {{11} \mathord{\left/ {\vphantom {{11} 8}} \right. } 8}}&{ - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{{{17} \mathord{\left/ {\vphantom {{17} 8}} \right. } 8}}&{ - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}} \\ {{1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{{{ - 17} \mathord{\left/ {\vphantom {{ - 17} 8}} \right. } 8}}&{ - {1 \mathord{\left/ {\vphantom {1 2}} \right. } 2}}&{{{91} \mathord{\left/ {\vphantom {{91} 8}} \right. } 8}} \end{array}} \right] |
Considering Eq (14) and Eq (15), the discontinuous projection can be defined as [11].
{\text{Pro}}{{\text{j}}_{{{\hat \alpha }_i}}}({ \cdot _i}) = \left\{ \begin{array}{l} 0, {\kern 1pt} {\kern 1pt} {\kern 1pt} { \cdot _i} > 0{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\text{and}}{\kern 1pt} {\kern 1pt} {{\hat \alpha }_i} = {\alpha _{i\max }} \hfill \\ 0, {\kern 1pt} {\kern 1pt} {\kern 1pt} { \cdot _i} < 0{\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\text{and}}{\kern 1pt} {\kern 1pt} {{\hat \alpha }_i} = {\alpha _{i\min }} \hfill \\ { \cdot _i}, {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\kern 1pt} {\text{otherwise}} \hfill \\ \end{array} \right. | (24) |
where \hat{\alpha} denote the estimate of \alpha and \tilde{\alpha} denote the estimate error, \tilde{\alpha} = \hat{\alpha}-\alpha, i = 1, 2, 3, 4.
Using the adaptation law as follow:
\dot{\hat{\alpha}} = \operatorname{Proj}_{\hat{\alpha}}(\Gamma \tau(t)), \hat{\alpha}(0) = \Omega_{\hat{\alpha}} | (25) |
where Γ is the diagonal positive definite adaptation rate matrix, τ is an adaptation function. For any adaption function τ, the adaptation Eq (25) satisfies follow [3]:
\begin{array}{l} \hat{\alpha} \in \Omega_{\dot{\alpha}} \triangleq\left\{\hat{\alpha}: \alpha_{\max } \leq \hat{\alpha} \leq \alpha_{\min }\right\} \end{array} | (26) |
\tilde{\alpha}^T\left[\Gamma^{-1} \operatorname{Proj}_{\hat{\alpha}}(\Gamma \tau)-\tau\right] \leq 0, \forall \tau | (27) |
Defining the following error variables
\left\{\begin{array}{l} z_1 = x_1-x_{1 d}, z_2 = \dot{z}_1+k_1 z_1 \\ z_3 = \dot{z}_2+k_2 z_2, r = \dot{z}_3+k_3 z_3 \end{array}\right. | (28) |
where x1d is the given trajectory; k1 k2 k3 are the positive feedback gain and r is the auxiliary error signal. Because r contains the differentiation of acceleration, it is considered to be unmeasurable in practice and only used for auxiliary design. According to Eq (28), r has the following expansion:
\begin{array}{l} r = \theta_1 u-\theta_2 x_{1 \mathrm{~d}}-\theta_3 \dot{x}_{1 \mathrm{~d}}-\theta_4 \ddot{x}_{\mathrm{ld}}+\Delta-\dddot{x}_{\mathrm{ld}}-\left(\theta_2-k_1 \theta_3-k_1^3+\theta_4 k_1^2\right) z_1- \\ \left(k_1^2+k_1 k_2-k_2^2 \theta_3-\theta_4 k_1-\theta_4 k_2\right) z_2+\left(k_1+k_2+k_3-\theta_4\right) z_3 \end{array} | (29) |
Dividing Eq (29) by θ1, and thus
\begin{array}{l} \alpha_1 r = u-\alpha_1 \dddot{x}_{1 \mathrm{~d}}-\alpha_2 x_{1 \mathrm{~d}}-\alpha_3 \dot{x}_{\mathrm{ld}}-\alpha_4 \ddot{x}_{1 \mathrm{~d}}+\alpha_1 \Delta-\left(\alpha_2-k_1 \alpha_3-\alpha_1 k_1^3+\alpha_4 k_1^2\right) z_1- \\ \left(\left(k_1^2+k_1 k_2\right) \alpha_1-k_2^2 \alpha_3-\alpha_4 k_1-\alpha_4 k_2\right) z_2+\left(\left(k_1+k_2+k_3\right) \alpha_1-\alpha_4\right) z_3 \end{array} | (30) |
where {\alpha _1} = {1 \mathord{\left/ {\vphantom {1 {{\theta _1}}}} \right. } {{\theta _1}}} , {\alpha _2} = {{{\theta _2}} \mathord{\left/ {\vphantom {{{\theta _2}} {{\theta _1}}}} \right. } {{\theta _1}}} , {\alpha _3} = {{{\theta _3}} \mathord{\left/ {\vphantom {{{\theta _3}} {{\theta _1}}}} \right. } {{\theta _1}}} , {\alpha _4} = {{{\theta _4}} \mathord{\left/ {\vphantom {{{\theta _4}} {{\theta _1}}}} \right. } {{\theta _1}}} .
The model-based controller is designed as follows:
\left\{ \begin{array}{l} u = {u_a} + {u_s}, {u_s} = \left( {{u_{s1}} + {u_{s2}}} \right), {u_{s1}} = - {k_3}{z_3} \hfill \\ {u_{s2}} = - {k_r}{z_3} + {k_r}{z_3}\left( 0 \right) - \int_o^t {\left[ {{k_r}{k_3}{z_3}\left( \tau \right) + \beta S\left( {{z_3}(\tau )} \right)} \right]{\text{d}}\tau } \hfill \\ {u_a} = {{\hat \alpha }_1}{{\dddot x}_{1{\text{d}}}} + {{\hat \alpha }_2}{x_{1{\text{d}}}} + {{\hat \alpha }_3}{{\dot x}_{1{\text{d}}}} + {{\hat \alpha }_4}{{\ddot x}_{1{\text{d}}}} - {{\hat \alpha }_1}\hat \Delta = {{\hat \alpha }^{\rm T}}{{\boldsymbol{\Phi }}_2} - {{\hat \alpha }_1}\hat \Delta \hfill \\ \end{array} \right. | (31) |
where {\hat \alpha = }{\left[{{{\hat \alpha }_1}, {{\hat \alpha }_2}, {{\hat \alpha }_3}, {{\hat \alpha }_4}} \right]^{\rm T}} represents the estimated value of {\alpha } = {\left[{{\alpha _1}, {\alpha _2}, {\alpha _3}, {\alpha _4}} \right]^{\rm T}} ; {k_r} > 0 is the gain of controller; \beta > 0 is the robust gain; {u_a} is the feedforward model compensation, us1 is the linear robust feedback term, us2 is the RISE control term, {{\boldsymbol{\varPhi }}_2} = {\left[{{{\dddot x}_{1{\text{d}}}}, {x_{1{\text{d}}}}, {{\dot x}_{1{\text{d}}}}, {{\ddot x}_{1{\text{d}}}}} \right]^{\rm T}} .
Substituting Eq (31) into Eq (30) and note that d = - {\hat \alpha _1}\hat \Delta + {\alpha _1}\Delta
{\alpha _1}r = {{\tilde{\boldsymbol{\alpha}} }^{\rm T}}{\boldsymbol{\varPhi }_2} + d - \left( {{k_3} + {k_r}} \right){z_3} + {k_r}{z_3}\left( 0 \right) - \int_o^t {\left[ {{k_r}{k_3}{z_3}\left( \tau \right) + \beta S\left( {{z_3}(\tau )} \right)} \right]{\text{d}}\tau } - A{z_1} - B{z_2} + C{z_3} | (32) |
{\tilde \alpha = \hat \alpha } - {\alpha } , Substitute Eq (25) into Eq (32), thus
{\alpha _1}\dot r = {\text{Pro}}{{\text{j}}_{\hat \alpha }}{(\Gamma \tau )^T}{{\boldsymbol{\varPhi} }_2} + {{\tilde{\boldsymbol{\alpha}} }^{\rm T}}{{\dot {\boldsymbol{\varPhi}} }_2} + \dot d - \left( {{k_3} + {k_r} - C} \right)r - \beta S\left( {{z_3}} \right) + A{k_1}{z_1} - \left( {A - B{k_2}} \right){z_2} - \left( {B + C{k_3}} \right){z_3} | (33) |
The overall structure of the designed control strategy is shown in Figure 2.
Lemma 1: Define variable L(t) as
L\left( t \right) = r\left[ {\dot d - \beta {\text{sign}}\left( {{z_3}} \right)} \right] | (34) |
Define auxiliary function as
P\left( t \right) = \beta \left| {{z_3}\left( 0 \right)} \right| - {z_3}\left( 0 \right)\dot d - \int_0^t {L\left( v \right)} {\text{d}}v | (35) |
According to [8], if the gain β satisfies the following inequality, then the auxiliary function P(t) is always positive definite.
\beta \geqslant {\delta _1} + \frac{1}{{{k_3}}}{\delta _2} | (36) |
Theorem 1: Using the adaptive law Eq (25), and adaptive function \tau = - r{{\boldsymbol{\dot \Phi }}_2} , and the robust gain β satisfies inequality Eq (36) as well as the feedback gains k1, k2, k3, kr are sufficient to ensure that the matrix {\Lambda } defined below is positive definite, the adaptive robust integral of the sign of the error controller Eq (31) can make all signals bounded in the closed-loop system, and the system obtains asymptotic output tracking, i.e., {z_1} \to 0 as t \to \infty .
{\boldsymbol{\varLambda}} = \left[ {\begin{array}{*{20}{c}} {{k_1}}&{ - \frac{1}{2}}&0&{ - \frac{1}{2}{k_5}} \\ { - \frac{1}{2}}&{{k_2}}&{ - \frac{1}{2}}&{ - \frac{1}{2}{k_6}} \\ 0&{ - \frac{1}{2}}&{{k_3}}&{ - \frac{1}{2}{k_{67}}} \\ { - \frac{1}{2}{k_5}}&{ - \frac{1}{2}{k_6}}&{ - \frac{1}{2}{k_7}}&{{k_4}} \end{array}} \right] | (37) |
where {k_4} = \max \left({{{{\dot {\boldsymbol{\varPhi}} }}_2}^{\rm T}\boldsymbol{\varGamma }{{\boldsymbol{\varPhi} }_2}} \right) + {k_3} + {k_r} - C , {k_5} = A{k_1} , {k_6} = - \left({A - B{k_2}} \right) , {k_7} = - \left({B + C{k_3}} \right) , \max \left(\cdot \right) represents the maximum value of the matrix.
Proof: see Appendix A.
The nominal value of the physical parameters of the valve controlled symmetrical hydraulic cylinder are shows in Table 1. The following controllers are compared by simulation to validate the effectiveness of the designed controller.
Parameter | Value | Unit | Parameter | Value | Unit |
mp | 0.76167 | kg | fv | 100 | N/(m/s) |
Ap | 2.5 × 10-4 | m2 | V10 | 1 × 10-3 | m3 |
k | 10900 | N/m | V20 | 1 × 10-3 | m3 |
βe | 2 × 108 | Pa | kt | 5.656 × 10-8 | m3/(s·V·N1/2) |
Ct | 1 × 10-13 | m3/(Pa·s) | q1 | 1 × 10-12 | m3/s |
q2 | 1 × 10-12 | m3/s | Cd | 0.7 | |
Cv | 1 | / | Wp | 5 | e-3 |
Δp | 821, 993 | / | α | 69 | ° |
K | 1000 |
(1) Controller I: ESO based ARISE This is the controller designed in this paper. The controller parameters are selected as: k1 = 200, k2 = 180, k3 = 0.08, kr = 0.012, β = 0.05, ω = 100. According to the nominal value of the parameters, the estimated boundary of unknown parameter α are given as: αmin = [5 × 10-5 0.03 2.5 8 × 10-3] and αmax = [7 × 10-5 0.07 3.5 20.2 × 10-3]. The initial estimates of α is set as \hat \alpha (0) = [6.5×10-5 0.04 3 8.2×10-3] and Γ is set as diag[1 × 10-15 2 × 10-2 0.001 5 × 10-10].
(2) Controller II: ARISE without ESO. Compared to the controller II, there is no ESO compensation term and the other parameters are same to controller I. That is only uα in Eq (18) is replaced as: {u_a} = {\hat \alpha _1}{\dddot x_{1{\text{d}}}} + {\hat \alpha _2}{x_{1{\text{d}}}} + {\hat \alpha _3}{\dot x_{1{\text{d}}}} + {\hat \alpha _4}{\ddot x_{1{\text{d}}}} = {\hat \alpha ^{\rm T}}{{\boldsymbol{\varPhi }}_2}
(3) Controller III: PI controller. The parameters are set as kP = 410 and kI = 10, which are the optimal solutions after repeated debugging.
(4) Controller IV: BP neutral network PID controller. The structure of the neutral network is 3-5-3, and the learning rate \eta = 0.28 , inertia coefficient α = 0.3.
The desired trajectories are designed as three cases: normal motion with the motion trajectory
{x_{\text{d}}}\left(t \right) = {{10\arctan \left[{\sin \left({\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} \mathord{\left/ {\vphantom {{10\arctan \left[{\sin \left({\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} {0.7854}}} \right. } {0.7854}} mm, fast level motion with the motion trajectory {x_{\text{d}}}\left(t \right) = {{10\arctan \left[{\sin \left({4\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} \mathord{\left/ {\vphantom {{10\arctan \left[{\sin \left({4\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} {0.7854}}} \right. } {0.7854}} mm and low-level motion with the motion trajectory {x_{\text{d}}}\left(t \right) = {{10\arctan \left[{\sin \left({0.2\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} \mathord{\left/ {\vphantom {{10\arctan \left[{\sin \left({0.2\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} {0.7854}}} \right. } {0.7854}} mm. The external disturbance is designed as f\left(t \right) = {{20\arctan \left[{\sin \left({0.8\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} \mathord{\left/ {\vphantom {{20\arctan \left[{\sin \left({0.8\pi t} \right)} \right]\left({1 - {e^{ - t}}} \right)} {0.7854}}} \right. } {0.7854}} N.
To compare the tracking responses of each controller quantitatively, three performance indices including maximum absolute value of the tracking error Me, average tracking error {\mu _{\text{e}}} , standard deviation of the tracking error σe, which were defined in are adopted to evaluate [20].
(1) Case I-normal level motion
The four controllers are tested for a normal motion trajectory xd(t) = 10arctan[sin(πt)] [1-exp(-t)]/0.7854 mm. The tracking performance are shown in Figures 4-6, the performance indices of the four controllers is shown in Table 2. From the Figures 4-6 and Table 1, it is obviously that the valve controlled symmetrical cylinder has the best tracking performance under the controller designed in this paper than other controllers. From Table 2, the amplitudes of steady-state tracking error of the controller III and controller IV are both about 0.6 mm, while controller I is about 0.003 mm and controller II is about 0.01 mm, it shows that the ARISE can deal with nonlinear and uncertainties and disturbance well but PI controller just has some robustness. By comparing the performance indices in Table 2 and tracking error in Figure 6 of controller I and controller II, it can be seen that controller I is better than controller II in all indices obviously, which indicates that the parameter adaptation in Figure 6 and ESO compensates for both parametric and uncertain lumped disturbance are effective. The control input u of controller I showed in Figure 3 is continuous and smooth, which makes it easy to implement in practice.
Indices | Me (mm) | μe (mm) | σe (mm) |
controller I | 0.00304985 | 0.000778897 | 0.00084379 |
controller II | 0.010971 | 0.00364635 | 0.00293964 |
controller III | 0.63097 | 0.299917 | 0.197092 |
controller IV | 0.630986 | 0.299916 | 0.197092 |
(2) Case II-low level motion
In this case, a low level reference trajectory xd(t) = 10arctan[sin(0.2πt)][1-exp(-t)]/0.7854 mm is tested. The results are shown in Figures 7-9 and the performance indices are listed in Table 3. From Figure 7 and Table 3, the amplitudes of steady-state tracking error of the controller III and controller IV are both about 0.076mm, while controller I and controller II both are about 9.5 × 10-5 mm, which shows that the ARISE can also deal with nonlinear and uncertainties and disturbance well in low level reference trajectory. By comparing the performance indices of controller I and controller II in Table 3, it can be seen that the maximum absolute value of the tracking error Me of the two controllers almost the same, but the average tracking error μe, and standard deviation of the tracking error σe of controller I are 6.536 × 10-6 mm and 8.833 × 10-6 mm respectively which are better than that of 2.87371 × 10-5 and 1.02822 × 10-5 of controller II obviously, which further validates the effectiveness of the desired parameter adaptation and ESO compensation. The parameter adaptation of controller I are omitted. The control input u of controller I showed in Figure 8 is continuous and there is slight high-frequency vibration. So the controller designed in this paper has the best tracking performance on low level reference trajectory than other controllers too.
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 9.508 × 10-5 | 6.536 × 10-6 | 8.833 × 10-6 |
controller II | 9.58784 × 10-5 | 2.87371 × 10-5 | 1.02822 × 10-5 |
controller III | 7.5716 × 10-2 | 3.14012 × 10-2 | 2.00589 × 10-2 |
controller IV | 7.57345 × 10-2 | 3.13926 × 10-2 | 2.00719 × 10-2 |
(3) Case III-fast level motion
A faster level reference trajectory xd(t) = 10arctan[sin(4πt)][1-exp(-t)]/0.7854 mm is tested in this case. The results are shown in Figured 10-12 and the performance indices are listed in Table 4. From Figure 12 and Table 4, the amplitudes of steady-state tracking error of the controller III and controller IV are both about 1.24 mm, while controller I 1.34 × 10-2 mm and controller II 4.20256 × 10-2 mm, which shows that the ARISE can better deal with nonlinear and uncertainties and disturbance in fast level reference trajectory than PI controller too. In addition, comparing the performance indices of controller I and controller II in Table 4, it can be seen that the maximum absolute value of the average tracking error μe of the controller I is 4.14 × 10-3 mm and the controller II is 1.63032 × 10-2 mm, the standard deviation of the tracking error σe of controller I are 5.53 ×10-3 mm and the controller II is 1.30355 × 10-2 mm. These further validate the effectiveness of the desired parameter adaptation and ESO compensation. The control input u of controller I showed in Figure 11 is continuous and smooth. The result verifies that the controller designed in this paper still has high tracking accuracy in tracking performance on fast level reference trajectory.
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 1.34 × 10-2 | 4.14 × 10-3 | 5.53 × 10-3 |
controller II | 4.20256 × 10-2 | 1.63032 × 10-2 | 1.30355 × 10-2 |
controller III | 1.238558 | 0.598359 | 0.391225 |
controller IV | 1.238657 | 0.598359 | 0.391224 |
In this paper, an ARISE with ESO controller is proposed for EHSS to address parametric uncertainties, uncertainty nonlinearities and unmolded disturbances. The proposed ARISE can compensate the dynamics uncertainties, thus guaranteeing asymptotic tracking and improving the adaptability and safety of EHSS. ESO can effectively estimate the state and lumped uncurtains. Simulation results shows that ARISE with ESO can obtain high tracking accuracy and better performance in tracking desired trajectory under all working conditions.
The authors declare there is no conflict of interest.
Define a Lyapunov function
V = \frac{1}{2} z_1^2+\frac{1}{2} z_2^2+\frac{1}{2} z_3^2+\frac{1}{2} \alpha_1 r^2+\frac{1}{2} \tilde{\boldsymbol{\alpha}}^{\mathrm{T}} \boldsymbol{\varGamma}^{-1} \tilde{\boldsymbol{\alpha}}+P | (A.1) |
It is Obvious that V is positive definite. The derivative of V is:
\dot{V} = z_1 \dot{z}_1+z_2 \dot{z}_2+z_3 \dot{z}_3+\alpha_1 r \dot{\eta}+\tilde{\boldsymbol{\alpha}}^{\mathrm{T}} \boldsymbol{\varGamma}^{-1} \dot{\hat{\boldsymbol{\alpha}}}+\dot{P} | (A.2) |
Substituting Eqs (25) and (28), L(t)and P(t), into Eq (A.2).
\begin{aligned} \dot{V} & = z_1\left(z_2-k_1 z_1\right)+z_2\left(z_3-k_2 z_2\right)+z_3\left(r-k_3 z_3\right)+ \\ & r\left\{-\dot{\boldsymbol{\varPhi}}_2^{\mathrm{T}} \boldsymbol{\varGamma} \boldsymbol{\Phi}_2 r+\tilde{\boldsymbol{\alpha}}^{\mathrm{T}} \dot{\boldsymbol{\varPhi}}_2+\dot{d}-\left(k_3+k_r-C\right) r-\beta \operatorname{sign}\left(z_3\right)+A k_1 z_1-\left(A-B k_2\right) z_2-\left(B+C k_3\right) z_3\right\} \\ & +\tilde{\boldsymbol{\alpha}}^T \boldsymbol{\varGamma}^{-1} \operatorname{Proj}_{\hat{\theta}}(\Gamma \tau)-r\left[\dot{d}-\beta \operatorname{sign}\left(z_3\right)\right] \\ & \leq z_1\left(z_2-k_1 z_1\right)+z_2\left(z_3-k_2 z_2\right)+z_3\left(r-k_3 z_3\right)+ \\ & r\left\{-\dot{\boldsymbol{\varPhi}}_2^{\mathrm{T}} \boldsymbol{\varGamma} \boldsymbol{\Phi}_2 r+\tilde{\boldsymbol{\alpha}}^{\mathrm{T}} \dot{\boldsymbol{\varPhi}}_2+\dot{d}-\left(k_3+k_r-C\right) r-\beta \operatorname{sign}\left(z_3\right)+A k_1 z_1-\left(A-B k_2\right) z_2-\left(B+C k_3\right) z_3\right\} \\ & +\tilde{\boldsymbol{\alpha}}^T \tau-r\left[\dot{d}-\beta \operatorname{sign}\left(z_3\right)\right] \\ & = -k_1 z_1^2-k_2 z_2^2-k_3 z_3^2-\left(\dot{\boldsymbol{\varPhi}}_2^{\mathrm{T}} \boldsymbol{\varGamma} \boldsymbol{\Phi}_2+k_3+k_r-C\right) r^2+z_1 z_2+z_2 z_3+z_3 r+\left(A k_1 z_1-A z_2\right) r+B k_2 z_2 r-\left(B+C k_3\right) z_3 r \\ & \triangleq-\boldsymbol{\eta}^{\mathrm{T}} \boldsymbol{\varLambda} \boldsymbol{\eta} \leq-\lambda_{\min }(\boldsymbol{\varLambda})\left(z_1^2+z_2^2+z_3^2+r^2\right) \triangleq-W \end{aligned} | (A.3) |
where \boldsymbol{\eta} = \left[z_1, z_2, z_3, r\right]^{\mathrm{T}}, \lambda \min (\boldsymbol{\varLambda}) is the minimum eigenvalue of matrix Λ, therefore V \in L_{\infty} and W \in L_2, so z1, z2, z3 and r are bounded. According to the assumptions 1 and 2, all states of the system are bounded so the actual control input u is bounded. According to Eqs (28) and (33), the derivative of W is bounded, so W is uniformly continuous. According to Barbarat's lemma, W→0 as t→∞, so the conclusion of the theorem 1 can be deduced, theorem 1 is proofed.
[1] |
J. Huang, B. Chen, B. Yao, W. He, ECG arrhythmia classification using STFT-based spectrogram and convolutional neural network, IEEE Access, 7 (2019), 92871–92880. https://doi.org/10.1109/ACCESS.2019.2928017 doi: 10.1109/ACCESS.2019.2928017
![]() |
[2] |
H. Chang, H. Zan, S. Zhang, B. Zhao, K. Zhang, Construction of cardiovascular information extraction corpus based on electronic medical records, Math. Biosci. Eng., 20 (2023), 13379–13397. https://doi.org/10.3934/mbe.2023596 doi: 10.3934/mbe.2023596
![]() |
[3] |
Q. Yao, R. Wang, X. Fan, J. Liu, Y. Li, Multi-class arrhythmia detection from 12-lead varied-length ECG using attention-based time-incremental convolutional neural network, Inf. Fusion, 53 (2020), 174–182. https://doi.org/10.1016/j.inffus.2019.06.024 doi: 10.1016/j.inffus.2019.06.024
![]() |
[4] |
L. C. M. Jr, J. F. Toole, H. S. Miller, Long-term EKG monitoring in patients with cerebrovascular insufficiency, Stroke, 7 (1976), 264–269. https://doi.org/10.1161/01.STR.7.3.264 doi: 10.1161/01.STR.7.3.264
![]() |
[5] |
M. Pittiruti, G. Scoppettuolo, A. L. Greca, A. Emoli, A. Brutti, I. Migliorini, et al., The EKG method for positioning the tips of PICCs: Results from two preliminary studies, J. Assoc. Vasc. Access, 13 (2008), 179–186. https://doi.org/10.2309/java.13-4-4 doi: 10.2309/java.13-4-4
![]() |
[6] |
E. K. Wang, L. Xi, R. P. Sun, F. Wang, L. Y. Pan, C. X. Cheng, et al., A new deep learning model for assisted diagnosis on electrocardiogram, Math. Biosci. Eng., 16 (2019), 2481–2491. https://doi.org/10.3934/mbe.2019124 doi: 10.3934/mbe.2019124
![]() |
[7] |
E. Svennberg, F. Tjong, A. Goette, N. Akoum, L. D. Biase, P. Bordachar, et al., How to use digital devices to detect and manage arrhythmias: An EHRA practical guide, Europace, 24 (2022), 979–1005. https://doi.org/10.1093/europace/euac038 doi: 10.1093/europace/euac038
![]() |
[8] | W. Midani, Z. Fki, M. BenAyed, Online anomaly detection in ECG signal using hierarchical temporal memory, in 2019 Fifth International Conference on Advances in Biomedical Engineering (ICABME), IEEE, (2019), 1–4. https://doi.org/10.1109/ICABME47164.2019.8940307 |
[9] | E. R. Adams, A. Choi, Using neural networks to predict cardiac arrhythmias, in 2012 IEEE International Conference on Systems, Man, and Cybernetics (SMC), IEEE, (2012), 402–407. https://doi.org/10.1109/ICSMC.2012.6377734 |
[10] |
W. Midani, W. Ouarda, M. B. Ayed, DeepArr: An investigative tool for arrhythmia detection using a contextual deep neural network from electrocardiograms (ECG) signals, Biomed. Signal Process. Control, 85 (2023), 104954. https://doi.org/10.1016/j.bspc.2023.104954 doi: 10.1016/j.bspc.2023.104954
![]() |
[11] |
S. M. P. Dinakarrao, A. Jantsch, M. Shafique, Computer-aided arrhythmia diagnosis with bio-signal processing: A survey of trends and techniques, ACM Comput. Surv., 52 (2019), 1–37. https://doi.org/10.1145/3297711 doi: 10.1145/3297711
![]() |
[12] |
J. J. Halford, Computerized epileptiform transient detection in the scalp electroencephalogram: Obstacles to progress and the example of computerized ECG interpretation, Clin. Neurophysiol., 120 (2009), 1909–1915. https://doi.org/10.1016/j.clinph.2009.08.007 doi: 10.1016/j.clinph.2009.08.007
![]() |
[13] |
M. Y. Ansari, Y. Yang, P. K. Meher, S. P. Dakua, Dense-PSP-UNet: A neural network for fast inference liver ultrasound segmentation, Comput. Biol. Med., 153 (2023), 106478. https://doi.org/10.1016/j.compbiomed.2022.106478 doi: 10.1016/j.compbiomed.2022.106478
![]() |
[14] |
Z. Akkus, J. Cai, A. Boonrod, A. Zeinoddini, A. D. Weston, K. A. Philbrick, et al., A survey of deep-learning applications in ultrasound: Artificial intelligence–powered ultrasound for improving clinical workflow, J. Am. Coll. Radiol., 16 (2019), 1318–1328. https://doi.org/10.1016/j.jacr.2019.06.004 doi: 10.1016/j.jacr.2019.06.004
![]() |
[15] |
F. A. Elhaj, N. Salim, A. R. Harris, T. T. Swee, T. Ahmed, Arrhythmia recognition and classification using combined linear and nonlinear features of ECG signals, Comput. Methods Programs Biomed., 127 (2016), 52–63. https://doi.org/10.1016/j.cmpb.2015.12.024 doi: 10.1016/j.cmpb.2015.12.024
![]() |
[16] |
Y. Kutlu, D. Kuntalp, Feature extraction for ECG heartbeats using higher order statistics of WPD coefficients, Comput. Methods Programs Biomed., 105 (2012), 257–267. https://doi.org/10.1016/j.cmpb.2011.10.002 doi: 10.1016/j.cmpb.2011.10.002
![]() |
[17] |
T. Li, M. Zhou, ECG classification using wavelet packet entropy and random forests, Entropy, 18 (2016), 285. https://doi.org/10.3390/e18080285 doi: 10.3390/e18080285
![]() |
[18] |
R. J. Martis, U. R. Acharya, C. M. Lim, K. M. Mandana, A. K. Ray, C. Chakraborty, Application of higher order cumulant features for cardiac health diagnosis using ECG signals, Int. J. Neural Syst., 23 (2013), 1350014. https://doi.org/10.1142/S0129065713500147 doi: 10.1142/S0129065713500147
![]() |
[19] |
R. J. Martis, U. R. Acharya, K. M. Mandana, A. K. Ray, C. Chakraborty, Application of principal component analysis to ECG signals for automated diagnosis of cardiac health, Expert Syst. Appl., 39 (2012), 11792–11800. https://doi.org/10.1016/j.eswa.2012.04.072 doi: 10.1016/j.eswa.2012.04.072
![]() |
[20] |
H. F. Huang, G. S. Hu, L. Zhu, Sparse representation-based heartbeat classification using independent component analysis, J. Med. Syst., 36 (2012), 1235–1247. https://doi.org/10.1007/s10916-010-9585-x doi: 10.1007/s10916-010-9585-x
![]() |
[21] |
U. R. Acharya, Y. Hagiwara, J. E. W. Koh, S. L. Oh, J. H. Tan, M. Adam, et al., Entropies for automated detection of coronary artery disease using ECG signals: A review, Biocybern. Biomed. Eng., 38 (2018), 373–384. https://doi.org/10.1016/j.bbe.2018.03.001 doi: 10.1016/j.bbe.2018.03.001
![]() |
[22] |
Ö. Yildirim, A novel wavelet sequence based on deep bidirectional LSTM network model for ECG signal classification, Comput. Biol. Med., 96 (2018), 189–202. https://doi.org/10.1016/j.compbiomed.2018.03.016 doi: 10.1016/j.compbiomed.2018.03.016
![]() |
[23] |
A. Khazaee, A. Ebrahimzadeh, Heart arrhythmia detection using support vector machines, Intell. Autom. Soft Comput., 19 (2013), 1–9. https://doi.org/10.1080/10798587.2013.771456 doi: 10.1080/10798587.2013.771456
![]() |
[24] | J. Park, K. Lee, K. Kang, Arrhythmia detection from heartbeat using k-nearest neighbor classifier, in 2013 IEEE International Conference on Bioinformatics and Biomedicine, IEEE, (2013), 15–22. https://doi.org/10.1109/BIBM.2013.6732594 |
[25] |
N. Maglaveras, T. Stamkopoulos, K. Diamantaras, C. Pappas, M. Strintzis, ECG pattern recognition and classification using non-linear transformations and neural networks: A review, Int. J. Med. Inform., 52 (1998), 191–208. https://doi.org/10.1016/S1386-5056(98)00138-5 doi: 10.1016/S1386-5056(98)00138-5
![]() |
[26] | M. Coşkun, Ö. YILDIRIM, U. Ayşegül, Y. Demir, An overview of popular deep learning methods, Eur. J. Tech., 7 (2017), 165–176. |
[27] |
U. Erdenebayar, H. Kim, J. U. Park, D. Kang, K. J. Lee, Automatic prediction of atrial fibrillation based on convolutional neural network using a short-term normal electrocardiogram signal, J. Korean Med. Sci., 34 (2019), e64. https://doi.org/10.3346/jkms.2019.34.e64 doi: 10.3346/jkms.2019.34.e64
![]() |
[28] |
K. S. Lee, S. Jung, Y. Gil, H. S. Son, Atrial fibrillation classification based on convolutional neural networks, BMC Med. Inform. Decis. Mak., 19 (2019), 1–6. https://doi.org/10.1186/s12911-019-0946-1 doi: 10.1186/s12911-019-0946-1
![]() |
[29] | C. Zhang, W. Liu, H. Ma, H. Fu, Siamese neural network based gait recognition for human identification, in 2016 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), IEEE, (2016), 2832–2836. https://doi.org/10.1109/ICASSP.2016.7472194 |
[30] |
Y. Ng, M. T. Liao, T. L. Chen, C. K. Lee, C. Y. Chou, W. Wang, Few-shot transfer learning for personalized atrial fibrillation detection using patient-based siamese network with single-lead ECG records, Artif. Intell. Med., 144 (2023), 102644. https://doi.org/10.1016/j.artmed.2023.102644 doi: 10.1016/j.artmed.2023.102644
![]() |
[31] |
L. Wang, X. Zhou, Detection of congestive heart failure based on LSTM-based deep network via short-term RR intervals, Sensors, 19 (2019), 1502. https://doi.org/10.3390/s19071502 doi: 10.3390/s19071502
![]() |
[32] | T. Y. Lin, P. Goyal, R. Girshick, K. He, P. Dollár, Focal loss for dense object detection, in 2017 IEEE International Conference on Computer Vision (ICCV), IEEE, (2017), 2980–2988. https://doi.org/10.1109/ICCV.2017.324 |
[33] |
T. F. Romdhane, M. A. Pr, Electrocardiogram heartbeat classification based on a deep convolutional neural network and focal loss, Comput. Biol. Med., 123 (2020), 103866. https://doi.org/10.1016/j.compbiomed.2020.103866 doi: 10.1016/j.compbiomed.2020.103866
![]() |
[34] |
Z. Zhu, W. Dai, Y. Hu, J. Li, Speech emotion recognition model based on Bi-GRU and focal loss, Pattern Recognit. Lett., 140 (2020), 358–365. https://doi.org/10.1016/j.patrec.2020.11.009 doi: 10.1016/j.patrec.2020.11.009
![]() |
[35] | R. Iikura, M. Okada, N. Mori, Improving bert with focal loss for paragraph segmentation of novels, in Distributed Computing and Artificial Intelligence, 17th International Conference, Springer, (2021), 21–30. https://doi.org/10.1007/978-3-030-53036-5_3 |
[36] |
G. Petmezas, K. Haris, L. Stefanopoulos, V. Kilintzis, A. Tzavelis, J. A. Rogers, et al., Automated atrial fibrillation detection using a hybrid CNN-LSTM network on imbalanced ECG datasets, Biomed. Signal Process. Control, 63 (2021), 102194. https://doi.org/10.1016/j.bspc.2020.102194 doi: 10.1016/j.bspc.2020.102194
![]() |
[37] |
M. Woźniak, M. Wieczorek, J. Siłka, BiLSTM deep neural network model for imbalanced medical data of IoT systems, Future Gener. Comput. Syst., 141 (2023), 489–499. https://doi.org/10.1016/j.future.2022.12.004 doi: 10.1016/j.future.2022.12.004
![]() |
[38] | S. V. Moravvej, S. J. Mousavirad, M. H. Moghadam, M. Saadatmand, An LSTM-based plagiarism detection via attention mechanism and a population-based approach for pre-training parameters with imbalanced classes, in Neural Information Processing, Springer, (2021), 690–701. https://doi.org/10.1007/978-3-030-92238-2_57 |
[39] | A. Jović, K. Brkić, N. Bogunović, A review of feature selection methods with applications, in 2015 38th International Convention on Information and Communication Technology, Electronics and Microelectronics (MIPRO), IEEE, (2015), 1200–1205. https://doi.org/10.1109/MIPRO.2015.7160458 |
[40] |
T. Emura, S. Matsui, H. Y. Chen, compound. Cox: Univariate feature selection and compound covariate for predicting survival, Comput. Methods Programs Biomed., 168 (2019), 21–37. https://doi.org/10.1016/j.cmpb.2018.10.020 doi: 10.1016/j.cmpb.2018.10.020
![]() |
[41] |
W. Lu, J. Li, J. Wang, L. Qin, A CNN-BiLSTM-AM method for stock price prediction, Neural Comput. Appl., 33 (2021), 4741–4753. https://doi.org/10.1007/s00521-020-05532-z doi: 10.1007/s00521-020-05532-z
![]() |
[42] |
Z. Niu, G. Zhong, H. Yu, A review on the attention mechanism of deep learning, Neurocomputing, 452 (2021), 48–62. https://doi.org/10.1016/j.neucom.2021.03.091 doi: 10.1016/j.neucom.2021.03.091
![]() |
[43] |
F. Murat, O. Yildirim, M. Talo, U. B. Baloglu, Y. Demir, U. R. Acharya, Application of deep learning techniques for heartbeats detection using ECG signals-analysis and review, Comput. Biol. Med., 120 (2020), 103726. https://doi.org/10.1016/j.compbiomed.2020.103726 doi: 10.1016/j.compbiomed.2020.103726
![]() |
[44] |
F. Qiao, B. Li, Y. Zhang, H. Guo, W. Li, S. Zhou, A fast and accurate recognition of ECG signals based on ELM-LRF and BLSTM algorithm, IEEE Access, 8 (2020), 71189–71198. https://doi.org/10.1109/ACCESS.2020.2987930 doi: 10.1109/ACCESS.2020.2987930
![]() |
[45] |
M. K. Ojha, S. Wadhwani, A. K. Wadhwani, A. Shukla, Automatic detection of arrhythmias from an ECG signal using an auto-encoder and SVM classifier, Phys. Eng. Sci. Med., 45 (2022), 665–674. https://doi.org/10.1007/s13246-022-01119-1 doi: 10.1007/s13246-022-01119-1
![]() |
[46] |
O. Yildirim, U. B. Baloglu, R. S. Tan, E. J. Ciaccio, U. R. Acharya, A new approach for arrhythmia classification using deep coded features and LSTM networks, Comput. Methods Programs Biomed., 176 (2019), 121–133. https://doi.org/10.1016/j.cmpb.2019.05.004 doi: 10.1016/j.cmpb.2019.05.004
![]() |
[47] |
M. Wu, Y. Lu, W. Yang, S. Y. Wong, A study on arrhythmia via ECG signal classification using the convolutional neural network, Front. Comput. Neurosci., 14 (2021), 564015. https://doi.org/10.3389/fncom.2020.564015 doi: 10.3389/fncom.2020.564015
![]() |
[48] |
S. L. Oh, E. Y. K. Ng, R. S. Tan, U. R. Acharya, Automated diagnosis of arrhythmia using combination of CNN and LSTM techniques with variable length heart beats, Comput. Biol. Med., 102 (2018), 278–287. https://doi.org/10.1016/j.compbiomed.2018.06.002 doi: 10.1016/j.compbiomed.2018.06.002
![]() |
[49] |
S. L. Oh, E. Y. K. Ng, R. S. Tan, U. R. Acharya, Automated beat-wise arrhythmia diagnosis using modified U-net on extended electrocardiographic recordings with heterogeneous arrhythmia types, Comput. Biol. Med., 105 (2019), 92–101. https://doi.org/10.1016/j.compbiomed.2018.12.012 doi: 10.1016/j.compbiomed.2018.12.012
![]() |
1. | Yiming Li, Zhongchao Zhang, Mingliang Bai, Guiqiu Song, Composite RISE control for vehicle-mounted servo system with unknown modeling uncertainties and unknown time-varying disturbances, 2024, 147, 00190578, 590, 10.1016/j.isatra.2024.02.017 | |
2. | Xiaohan Yang, Guozhen Cheng, Yinghao Cui, Jie Hang, ESO-based robust adaptive control for dual closed-loop fuel control system in aeroengine, 2024, 144, 09670661, 105835, 10.1016/j.conengprac.2023.105835 | |
3. | Haifang Zhong, Kailei Liu, Hongbin Qiang, Jing Yang, Shaopeng Kang, Model reference adaptive control of electro-hydraulic servo system based on RBF neural network and nonlinear disturbance observer, 2024, 0959-6518, 10.1177/09596518241277714 |
Parameter | Value | Unit | Parameter | Value | Unit |
mp | 0.76167 | kg | fv | 100 | N/(m/s) |
Ap | 2.5 × 10-4 | m2 | V10 | 1 × 10-3 | m3 |
k | 10900 | N/m | V20 | 1 × 10-3 | m3 |
βe | 2 × 108 | Pa | kt | 5.656 × 10-8 | m3/(s·V·N1/2) |
Ct | 1 × 10-13 | m3/(Pa·s) | q1 | 1 × 10-12 | m3/s |
q2 | 1 × 10-12 | m3/s | Cd | 0.7 | |
Cv | 1 | / | Wp | 5 | e-3 |
Δp | 821, 993 | / | α | 69 | ° |
K | 1000 |
Indices | Me (mm) | μe (mm) | σe (mm) |
controller I | 0.00304985 | 0.000778897 | 0.00084379 |
controller II | 0.010971 | 0.00364635 | 0.00293964 |
controller III | 0.63097 | 0.299917 | 0.197092 |
controller IV | 0.630986 | 0.299916 | 0.197092 |
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 9.508 × 10-5 | 6.536 × 10-6 | 8.833 × 10-6 |
controller II | 9.58784 × 10-5 | 2.87371 × 10-5 | 1.02822 × 10-5 |
controller III | 7.5716 × 10-2 | 3.14012 × 10-2 | 2.00589 × 10-2 |
controller IV | 7.57345 × 10-2 | 3.13926 × 10-2 | 2.00719 × 10-2 |
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 1.34 × 10-2 | 4.14 × 10-3 | 5.53 × 10-3 |
controller II | 4.20256 × 10-2 | 1.63032 × 10-2 | 1.30355 × 10-2 |
controller III | 1.238558 | 0.598359 | 0.391225 |
controller IV | 1.238657 | 0.598359 | 0.391224 |
Parameter | Value | Unit | Parameter | Value | Unit |
mp | 0.76167 | kg | fv | 100 | N/(m/s) |
Ap | 2.5 × 10-4 | m2 | V10 | 1 × 10-3 | m3 |
k | 10900 | N/m | V20 | 1 × 10-3 | m3 |
βe | 2 × 108 | Pa | kt | 5.656 × 10-8 | m3/(s·V·N1/2) |
Ct | 1 × 10-13 | m3/(Pa·s) | q1 | 1 × 10-12 | m3/s |
q2 | 1 × 10-12 | m3/s | Cd | 0.7 | |
Cv | 1 | / | Wp | 5 | e-3 |
Δp | 821, 993 | / | α | 69 | ° |
K | 1000 |
Indices | Me (mm) | μe (mm) | σe (mm) |
controller I | 0.00304985 | 0.000778897 | 0.00084379 |
controller II | 0.010971 | 0.00364635 | 0.00293964 |
controller III | 0.63097 | 0.299917 | 0.197092 |
controller IV | 0.630986 | 0.299916 | 0.197092 |
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 9.508 × 10-5 | 6.536 × 10-6 | 8.833 × 10-6 |
controller II | 9.58784 × 10-5 | 2.87371 × 10-5 | 1.02822 × 10-5 |
controller III | 7.5716 × 10-2 | 3.14012 × 10-2 | 2.00589 × 10-2 |
controller IV | 7.57345 × 10-2 | 3.13926 × 10-2 | 2.00719 × 10-2 |
Indices | Me(mm) | μe(mm) | σe(mm) |
controller I | 1.34 × 10-2 | 4.14 × 10-3 | 5.53 × 10-3 |
controller II | 4.20256 × 10-2 | 1.63032 × 10-2 | 1.30355 × 10-2 |
controller III | 1.238558 | 0.598359 | 0.391225 |
controller IV | 1.238657 | 0.598359 | 0.391224 |