A noise-immune reinforcement learning method for early diagnosis of neuropsychiatric systemic lupus erythematosus

Guanru Tan; Boyu Huang; Zhihan Cui; Haowen Dou; Shiqiang Zheng; Teng Zhou; Guanru Tan; Boyu Huang; Zhihan Cui; Haowen Dou; Shiqiang Zheng; Teng Zhou

doi:10.3934/mbe.2022104

Mathematical Biosciences and Engineering

2022, Volume 19, Issue 3: 2219-2239. doi: 10.3934/mbe.2022104

Previous Article Next Article

Research article Special Issues

A noise-immune reinforcement learning method for early diagnosis of neuropsychiatric systemic lupus erythematosus

1.
Department of Computer Science, Shantou University, Shantou 515063, China
2.
Key Laboratory of Intelligent Manufacturing Technology, Shantou University, Ministry of Education, Shantou 515063, China

Academic Editor: Alessio Martino

Received: 22 September 2021 Revised: 25 November 2021 Accepted: 23 December 2021 Published: 04 January 2022

The neuropsychiatric systemic lupus erythematosus (NPSLE), a severe disease that can damage the heart, liver, kidney, and other vital organs, often involves the central nervous system and even leads to death. Magnetic resonance spectroscopy (MRS) is a brain functional imaging technology that can detect the concentration of metabolites in organs and tissues non-invasively. However, the performance of early diagnosis of NPSLE through conventional MRS analysis is still unsatisfactory. In this paper, we propose a novel method based on genetic algorithm (GA) and multi-agent reinforcement learning (MARL) to improve the performance of the NPSLE diagnosis model. Firstly, the proton magnetic resonance spectroscopy ( $^{1}$ H-MRS) data from 23 NPSLE patients and 16 age-matched healthy controls (HC) were standardized before training. Secondly, we adopt MARL by assigning an agent to each feature to select the optimal feature subset. Thirdly, the parameter of SVM is optimized by GA. Our experiment shows that the SVM classifier optimized by feature selection and parameter optimization achieves 94.9% accuracy, 91.3% sensitivity, 100% specificity and 0.87 cross-validation score, which is the best score compared with other state-of-the-art machine learning algorithms. Furthermore, our method is even better than other dimension reduction ones, such as SVM based on principal component analysis (PCA) and variational autoencoder (VAE). By analyzing the metabolites obtained by MRS, we believe that this method can provide a reliable classification result for doctors and can be effectively used for the early diagnosis of this disease.

Keywords:

neuropsychiatric systemic lupus erythematosus,
magnetic resonance spectroscopy,
high dimensional data,
limited samples,
genetic algorithms,
reinforcement learning

Citation: Guanru Tan, Boyu Huang, Zhihan Cui, Haowen Dou, Shiqiang Zheng, Teng Zhou. A noise-immune reinforcement learning method for early diagnosis of neuropsychiatric systemic lupus erythematosus[J]. Mathematical Biosciences and Engineering, 2022, 19(3): 2219-2239. doi: 10.3934/mbe.2022104

Related Papers:

[1]	Aziz Belmiloudi . Time-varying delays in electrophysiological wave propagation along cardiac tissue and minimax control problems associated with uncertain bidomain type models. AIMS Mathematics, 2019, 4(3): 928-983. doi: 10.3934/math.2019.3.928
[2]	Zuliang Lu, Xiankui Wu, Fei Huang, Fei Cai, Chunjuan Hou, Yin Yang . Convergence and quasi-optimality based on an adaptive finite element method for the bilinear optimal control problem. AIMS Mathematics, 2021, 6(9): 9510-9535. doi: 10.3934/math.2021553
[3]	Zahra Pirouzeh, Mohammad Hadi Noori Skandari, Kamele Nassiri Pirbazari, Stanford Shateyi . A pseudo-spectral approach for optimal control problems of variable-order fractional integro-differential equations. AIMS Mathematics, 2024, 9(9): 23692-23710. doi: 10.3934/math.20241151
[4]	Xin Yi, Rong Liu . An age-dependent hybrid system for optimal contraception control of vermin. AIMS Mathematics, 2025, 10(2): 2619-2633. doi: 10.3934/math.2025122
[5]	Yuanyuan Cheng, Yuan Li . A novel event-triggered constrained control for nonlinear discrete-time systems. AIMS Mathematics, 2023, 8(9): 20530-20545. doi: 10.3934/math.20231046
[6]	Xiang Wu, Yuzhou Hou, Kanjian Zhang . Optimal feedback control for a class of fed-batch fermentation processes using switched dynamical system approach. AIMS Mathematics, 2022, 7(5): 9206-9231. doi: 10.3934/math.2022510
[7]	Woocheol Choi, Young-Pil Choi . A sharp error analysis for the DG method of optimal control problems. AIMS Mathematics, 2022, 7(5): 9117-9155. doi: 10.3934/math.2022506
[8]	Qian Li, Zhenghong Jin, Linyan Qiao, Aichun Du, Gang Liu . Distributed optimization of nonlinear singularly perturbed multi-agent systems via a small-gain approach and sliding mode control. AIMS Mathematics, 2024, 9(8): 20865-20886. doi: 10.3934/math.20241015
[9]	Tainian Zhang, Zhixue Luo . Optimal harvesting for a periodic competing system with size structure in a polluted environment. AIMS Mathematics, 2022, 7(8): 14696-14717. doi: 10.3934/math.2022808
[10]	Asaf Khan, Gul Zaman, Roman Ullah, Nawazish Naveed . Optimal control strategies for a heroin epidemic model with age-dependent susceptibility and recovery-age. AIMS Mathematics, 2021, 6(2): 1377-1394. doi: 10.3934/math.2021086

Abstract

1. Introduction

In structural mechanics, particularly in the field of elasticity, there are equations that describe the behavior of thin plates under loads. These equations are often partial differential equations that govern the displacement of a thin plate. The plate equation depends on factors like material properties, geometry, and boundary conditions. In the context of geophysics, "plate equation" could refer to the equations that describe the movement and interaction of tectonic plates on the Earth's surface. Plate tectonics is a theory that explains the movement of the Earth's lithosphere (the rigid outer layer of the Earth) on the more fluid asthenosphere beneath it. In mathematics, specifically in the field of differential equations, the term "plate equation" might be used to refer to certain types of equations. For instance, in polar coordinates, Laplace's equation takes a specific form that is sometimes informally referred to as the "plate equation". Plate equation models are hyperbolic systems that arise in several areas in real-life problems, (see, for instance, Kizilova et al.^[1], Lasiecka et al. ^[2] and Huang et al.^[3]). The theory of plates is the mathematical formulation of the mechanics of flat plates. It is defined as flat structural components with a low thickness compared to plane dimensions. The advantage of the theory of plates comes from the disparity of the length scale to reduce the problem of the mechanics of three-dimensional solids to a two-dimensional problem. The purpose of this theory is to compute the stresses and deformation in a loaded plate. The equation of plates results from the composition of different subsets of plates: The equilibrium equations, constitutive, kinematic, and force resultant, ^[4,5,6].

Following this, there are a wide number of works devoted to the analysis and control of the academic model of hyperbolic systems, the so-called plate equations, For example, the exact and the approximate controllability of thermoelastic plates given by Eller et al. ^[7] and Lagnese and Lions in ^[8] treated the control of thin plates and Lasiecka in ^[9] considered the controllability of the Kirchoff plate. Zuazua ^[10] treated the exact controllability for semi-linear wave equations. Recently many problems involving a plate equations were considered by researchers. Let us cite as examples the stabilization of the damped plate equation under general boundary conditions by Rousseau an Zongo ^[11]; the null controllability for a structurally damped stochastic plate equation studied by Zhao ^[12], Huang et al. ^[13] considerrd a thermal equation of state for zoisite: Implications for the transportation of water into the upper mantle and the high-velocity anomaly in the Farallon plate. Kaplunov et al. ^[14] discussed the asymptotic derivation of 2D dynamic equations of motion for transversely inhomogeneous elastic plates. Hyperbolic systems have recently continued to be of interest to researchers and many results have been obtained. We mention here the work of Fu et al. ^[15] which discusses a class of mixed hyperbolic systems using iterative learning control. Otherwise, for a class of one-dimension linear wave equations, Hamidaoui et al. stated in ^[16] an iterative learning control. Without forgetting that for a class of second-order nonlinear systems Tao et al. proposed an adaptive control based on an disturbance observer in ^[17] to improve the tracking performance and compensation. In addition to these works, the optimal control of the Kirchoff plate using bilinear control was considered by Bradly and Lenhart in ^[18], and Bradly et al. in ^[19]. In fact, in this work we will talk about a bilinear plate equation and we must cite the paper of Zine ^[20] which considers a bilinear hyperbolic system using the Riccati equation. Zine and Ould Sidi ^[19,22] that introduced the notion of partial optimal control of bilinear hyperbolic systems. Li et al. ^[23] give an iterative method for a class of bilinear systems. Liu, et al. ^[24] extended a gradient-based iterative algorithm for bilinear state-space systems with moving average noises by using the filtering technique. Furthermore, flow analysis of hyperbolic systems refers to the problems dealing with the analysis of the flow state on the system domain. We can refer to the work of Benhadid et al. on the flow observability of linear and semilinear systems ^[25], Bourray et al. on treating the controllability flow of linear hyperbolic systems ^[26] and the flow optimal control of bilinear parabolic systems are considered by Ould Sidi and Ould Beinane on ^[27,28].

For the motivation the results proposed in this paper open a wide range of applications. We cite the problem of iterative identification methods for plate bilinear systems ^[23], as well as the problem of the extended flow-based iterative algorithm for a plate systems ^[24].

This paper studies the optimal control problem governed by an infinite dimensional bilinear plate equation. The objective is to command the flow state of the bilinear plate equation to the desired flow using different types of bounded feedback. We show how one can transfer the flow of a plate equation close to the desired profile using optimization techniques and adjoint problems. As an application, we solve the partial flow control problem governed by a plate equation. The results open a wide way of applications in fractional systems. We began in section two by the well-posedness of our problem. In section three, we prove the existence of an optimal control solution of (2.3). In section four, we state the characterization of the optimal control. In section five we debate the case of time bilinear optimal control. Section six, proposes a method for solving the flow partial optimal control problem governed by a plate equation.

2. Well-posedness of the problem

Consider $\Theta$ an open bounded domain of $I\!\!R^2$ with $C^2$ boundary, for a time $m$ , and $\Gamma = \partial\Theta \times (0, m)$ . The control space time set is such that

$\begin{equation} Q \in U_{p} = \{ Q\in L^{\infty}([0,m];L^{\infty}(\Theta)){\mbox{ such that }} -p\leq Q(t) \leq p\}, \end{equation}$

(2.1)

with $p$ as a positive constant. Let the plate bilinear equation be described by the following system

$\begin{equation} \left \{ \begin{array}{lll} {\frac{{\partial }^2 {u} }{{\partial t}^2} } + \Delta^{2} u = Q(t)u_{t}, & & (0,m) \times \Theta, \\ u(x,0) = u_0(x) ,\; {\frac{{\partial } u }{{\partial t}} } (x,0) = u_1(x), & & \Theta,\\ u = \frac{\partial u}{\partial \nu} = 0, & & \Gamma, \end{array} \right. \end{equation}$

(2.2)

where $u_{t} = \frac{\partial u}{\partial t}$ is the velocity. The state space is $H_{0}^{2}(\Theta)\times L^{2}(\Theta)$ , (see Lions and Magenes ^[29] and Brezis ^[30]). We deduce the existence and uniqueness of the solution for (2.2) using the classical results of Pazy ^[31]. For $\lambda > 0$ , we define $\nabla u$ as the flow control problem governed by the bilinear plate equation (2.2) as the following:

$\begin{equation} \min\limits_{Q\ \in U_{p} }C_{\lambda}(Q), \end{equation}$

(2.3)

where $C_{\lambda}$ , is the flow penalizing cost defined by

$\begin{align} C_{\lambda}(Q) = \frac{1}{2}\Big\|\nabla u-u^{d}\Big\|^{2}_{(L^2(0,m;L^2(\Theta)))^{n}} + \frac{\lambda}{2}\int_{0}^{m} \int_{\Theta} Q^2(x,t)dxdt\\ = \frac{1}{2}\sum\limits_{i = 1}^{n} \Big\|\frac{\partial u}{\partial x_{i}} - u_{i}^{d}\Big\|^{2}_{L^2(0,m;L^2(\Theta))} + \frac{\lambda}{2}\int_{0}^{m} \int_{\Theta} Q^2(x,t)dxdt, \end{align}$

(2.4)

where $u^d = (u_{1}^{d}, ....u_{n}^{d})$ is the flow target in $L^2(0, m;L^2(\Theta))$ . One of the important motivations when considering the problem (2.3) is the isolation problems, where the control is maintained to reduce the flow temperature on the surface of a thin plate (see El Jai et al. ^[32]).

3. Existence of solution

Lemma 3.1. If $(u_{0}, u_{1}) \in H_{0}^{2}(\Theta)\times L^{2}(\Theta)$ and $Q \in U_{p}$ , then the solution $(u, u_{t})$ of (2.2) satisfies the following estimate:

$\Big\| (u, u_{t}) \Big\|_{{\cal C}(0,m; H_{0}^{2}(\Theta)\times L^{2}(\Theta))} \leq T (1+ \eta m)e^{\eta K m},$

where $T = \Big\| (u_{0}, u_{1}) \Big\|_{ H_{0}^{2}(\Theta)\times L^{2}(\Theta)}$ and $K$ is a positive constant ^[18,19].

Using the above Lemma 3.1, we prove the existence of an optimal control solution of (2.3).

Theorem 3.1. $(u^{*}, Q^*)\in {\cal C}([0, m];\; \; H_{0}^{2}(\Theta)\times U_{p})$ , is the solution of (2.3), where $u^{*}$ is the output of (2.2) and $Q^*$ is the optimal control function.

Proof. Consider the minimizing sequence $(Q_{n})_{n}$ in $U_{p}$ verifying

$C^{*} = \lim\limits_{n\rightarrow +\infty}C_{\lambda}(Q_n) = \inf\limits_{Q\in L^{\infty}(0,m; L^{\infty}(\Theta))} C_{\lambda}(Q).$

We choose $\bar{u}^{n} = (u^{n}, \frac{\partial u^{n}}{\partial t})$ to be the corresponding state of Eq (2.2). Using Lemma 3.1, we deduce

$\begin{equation} \Big\| u^{n}(x,t) \Big\|_{H_{0}^{2}(\Theta)}^{2} + \Big\| u_{t}^{n} (x,t) \Big\|_{ L^{2}(\Theta)}^{2} \leq T_{1} e^{\eta K m} {\mbox{ for }} 0\leq t\leq m {\mbox{ and }} T_{1}\in I\!\!R^{+}. \end{equation}$

(3.1)

Furthermore, system (2.2) gives

$\Big\| u^{n}_{tt}(x,t) \Big\|_{H^{-2}(\Theta)}^{2} \leq T_{2} \Big\| u_{t}^{n} (x,t) \Big\|_{ L^{2}(\Theta)}^{2} {\mbox{ with }} T_{2}\in I\!\!R^{+}.$

Then easily from (3.1), we have

$\begin{equation} \Big\| u^{n}_{tt}(x,t) \Big\|_{H^{-2}(\Theta)}^{2} \leq T_{3} e^{\eta K m} {\mbox{ for }} 0\leq t\leq m {\mbox{ and }} T_{3}\in I\!\!R^{+}. \end{equation}$

(3.2)

Using (3.1) and (3.2), we have the following weak convergence:

$\begin{equation} \begin{array}{lll} Q_{n} \rightharpoonup Q^{*}, & & L^{2}(0,m ; L^{2}(\Theta)),\\ u^{n} \rightharpoonup u^{*}, & & L^{\infty}(0,m ; H_{0}^{2}(\Theta)),\\ u_{t}^{n} \rightharpoonup u^{*}_{t}, & & L^{\infty}(0,m ; L^{2}(\Theta)),\\ u_{tt}^{n} \rightharpoonup u^{*}_{tt}, & & L^{\infty}(0,m ; H^{-2}(\Theta)),\\ \end{array} \end{equation}$

(3.3)

From the first convergence property of (3.3) with a control sequence $Q_{n} \in U_{p}$ , easily one can deduce that $Q^{*} \in U_{p}$ ^[20].

In addition, the mild solution of (2.2) verifies

$\begin{align} \int_{0}^{m} u^{n}_{tt}f(t)dt+ \int_{0}^{m} \int_{\Theta}\Delta u^{n} \Delta f(t) dx dt = \int_{0}^{m} Q^{*}u_{t}^{n} f(t)dt, \forall f \in H_{0}^{2}(\Theta). \end{align}$

(3.4)

Using (3.3) and (3.4), we deduce that

$\begin{align} \int_{0}^{m} u^{*}_{tt}f(t)dt+ \int_{0}^{m} \int_{\Theta}\Delta u^{*} \Delta f(t) dx dt = \int_{0}^{m} Q^{*}u_{t}^{*} f(t)dt, \forall f \in H_{0}^{2}(\Theta), \end{align}$

(3.5)

which implies that $u^{*} = u(Q_{*})$ is the output of (2.2) with command function $Q^{*}$ .

Fatou's lemma and the lower semi-continuous property of the cost $C_{\lambda}$ show that

$\begin{align} C_{\lambda}(Q^{*}) \leq \frac{1}{2}\lim\limits_{k\rightarrow +\infty}\Big\|\nabla u^{k}-u^{d}\Big\|^{2}_{(L^2(0,m;L^2(\Theta)))^{n}} + \frac{\lambda}{2}\lim\limits_{k\rightarrow +\infty} \int_{0}^{m} \int_{\Theta} Q_{k}^2(x,t)dxdt\\ \leq \liminf\limits_{k\rightarrow +\infty} C_{\lambda}(Q_{n})\\ = \inf\limits_{Q \in U_{p}} C_{\lambda}(Q), \end{align}$

(3.6)

which allows us to conclude that $Q^{*}$ is the solution of problem (2.3). □

4. Characterization of solution

We devote this section to establish a characterization of solutions to the flow optimal control problem (2.3).

Let the system

$\begin{equation} \left \{ \begin{array}{lll} {\frac{{\partial^{2} } v }{{\partial t^{2}}} } = -\Delta^{2}v(x,t)+Q(x,t) v_{t} + d(x,t) v_{t}, & & (0, m)\times \Theta,\\ v(x,0) = v_{t}(x,0) = v_0(x) = 0, & & \Theta, \\ v = \frac{\partial v}{\partial \nu } = 0, & & \Gamma, \end{array} \right. \end{equation}$

(4.1)

with $d \in L^{\infty}(0, m; L^{\infty}(\Theta))$ verify $Q+\delta d \in U_{p}, \quad \forall \delta > 0$ is a small constant. The functional defined by $Q \in U_{p}\mapsto \bar{u}(Q) = (u, u_{t}) \in {\cal C}(0, m; H_{0}^{2}(\Theta)\times L^{2}(\Theta))$ is differentiable and its differential $\overline{v} = (v, v_{t})$ is the solution of (4.1) ^[21].

The next lemma characterizes the differential of our flow cost functional $C_{\lambda}(Q)$ with respect to the control function $Q$ .

Lemma 4.1. Let $Q \in U_{p}$ and the differential of $C_{\lambda}(Q)$ can be written as the following:

$\begin{align} \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} = \sum\limits_{i = 1}^{n} \int _{\Theta}\int_{0}^{m} \frac{\partial v(x,t)}{\partial x_{i}}(\frac{\partial u}{\partial x_{i}} - u_{i}^{d} )dtdx+ \varepsilon\int_{\Theta}\int_{0}^{m} d Q dt dx.\\ \end{align}$

(4.2)

Proof. Consider the cost $C_{\lambda}(Q)$ defined by (2.4), which is

$\begin{align} C_{\lambda}(Q) = \frac{1}{2} \sum\limits_{i = 1}^{n}\int_{\Theta}\int_{0}^{m}(\frac{\partial u}{\partial x_{i}} - u_{i}^{d} )^2 dt dx + \frac{\lambda}{2} \int_{\Theta}\int^{m}_{0} Q^2(t) dt dx. \end{align}$

(4.3)

Put $u_{k} = z(Q+k d)$ , $u = u(Q)$ , and using (4.3), we have

$\begin{align} \lim\limits_{k \longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} = \lim\limits_{\beta\longrightarrow 0}\sum\limits_{i = 1}^{n} \frac{1}{2}\int_{\Theta}\int_{0}^{m}\frac{( \frac{\partial u_{k}}{\partial x_{i}} - u_i^d)^{2}-( \frac{\partial u}{\partial x_{i}}-u_i^d)^{2}}{k}dt dx\\ + \lim\limits_{k \longrightarrow 0}\frac{\lambda}{2}\int_{\Theta} \int_{0}^{m}\frac{(Q+k d)^2-Q^2}{k}(t)dt dx. \end{align}$

(4.4)

Consequently

$\begin{align} \lim\limits_{k \longrightarrow 0} \frac{C_{\lambda}(Q +kd)-C_{\lambda}(Q )}{k} \\ = \lim\limits_{k \longrightarrow 0}\sum\limits_{i = 1}^{n}\frac{1}{2}\int_{\Theta}\int^{m}_{0}\frac{( \frac{\partial u_{k}}{\partial x_{i}}- \frac{\partial u}{\partial x_{i}})}{k}( \frac{\partial u_{k} }{\partial x_{i}}+ \frac{\partial u}{\partial x_{i}}-2u_i^d) dt dx \\ + \lim\limits_{k \longrightarrow 0} \int_{\Theta}\int_{0}^{m}(\lambda d Q+ k \lambda d^2)dt dx\\ = \sum\limits_{i = 1}^{n}\int_\Theta \int_{0}^{m}\frac{\partial v(x,t)}{\partial x_{i}}( \frac{\partial u(x,t)}{\partial x_{i}}-u_i^d)dt dx+\int_\Theta \int_{0}^{m}\lambda d Q dt dx. \end{align}$

(4.5)

□ We define the following family of adjoint equations for system (4.1)

$\begin{equation} \left \{ \begin{array}{lll} {\frac{{\partial^{2} } {w_{i}}}{{\partial t^{2}}} } +\Delta^{2} w_{i} = Q^*(x,t) (w_{i})_{t}+( \frac{\partial u}{\partial x_{i}} - u_{i}^{d}), & & (0, m)\times\Theta, \\ w_{i}(x,m) = (w_{i})_{t}(x,m) = 0, & & \Theta, \\ w_{i} = \frac{\partial w_{i}}{\partial \nu } = 0, & & \Gamma. \end{array} \right. \end{equation}$

(4.6)

Such systems allow us to characterize the optimal control solution of (2.3).

Theorem 4.1. Consider $Q \in U_{p}$ , and $u = u (Q)$ its corresponding state space solution of (2.2), then the control solution of (2.3) is

$\begin{equation} Q(x,t) = max(-p, min(\frac{-1}{\lambda}(u_{t})(\sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}}),p)), \end{equation}$

(4.7)

where $w = (w_{1}....w_{n})$ with $w_{i} \in C ([ 0, T ]; H^{2}_{0}(\Theta))$ is the unique solution of (4.6).

Proof. Choose $d\in U_{p}$ such that $Q + k d \in U_{p}$ with $k > 0$ . The minimum of $C_{\lambda}$ is realized when the control $Q$ , verifies the following condition:

$\begin{equation} 0 \leq \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q )}{k}. \end{equation}$

(4.8)

Consequently, Lemma 4.1 gives

$\begin{align} 0 \leq \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} \\ = \sum\limits_{i = 1}^{n}\int_\Theta \int_{0}^{m}\frac{\partial v(x,t)}{\partial x_{i}}( \frac{\partial u(x,t)}{\partial x_{i}}-u_i^d)dt dx+\int_\Theta \int_{0}^{m}\lambda d Q dt dx.\\\\ \end{align}$

(4.9)

Substitute by equation (4.6) and we find

$\begin{align} 0 \leq \sum\limits_{i = 1}^{n}\int_\Theta \int_{0}^{m}\frac{\partial v(x,t)}{\partial x_{i}}( {\frac{{\partial^{2} } {w_{i}}(x,t) }{{\partial t^{2}}} } +\Delta^{2} w_{i}(x,t)-Q(x,t)( w_{i})_{t}(x,t))dt dx+\int_\Theta \int_{0}^{m}\lambda d Q dt dx\\ \\ = \sum\limits_{i = 1}^{n}\int_\Theta \int_{0}^{m}\frac{\partial }{\partial x_{i}}( {\frac{{\partial^{2} } v}{{\partial t^{2}}} } +\Delta^{2} v-Q(x,t) v_{t})w_{i}(x,t)dt dx+\int_\Theta \int_{0}^{m}\lambda d Qdt dx\\ \\ = \sum\limits_{i = 1}^{n}\int_\Theta \int_{0}^{m}\frac{\partial }{\partial x_{i}}(d(x,t) u_{t})w_{i}dt dx+\int_\Theta \int_{0}^{m}\lambda dQ dt dx\\ \\ = \int_\Theta \int_{0}^{m}d(x,t)[u_{t}(\sum\limits_{i = 1}^{n}\frac{\partial w_{i}(x,t)}{\partial x_{i}})+\lambda Q dt dx].\\ \\ \end{align}$

(4.10)

It is known that if $d = d(t)$ in a chosen function with $Q+k d \in U_{p}$ , using Bang-Bang control properties, one can conclude that

$\begin{equation} Q(x,t) = max(-p, min(\frac{-u_{t}}{\lambda}(\sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}}),p)) = max(-p, min( \frac{-u_{t}}{\lambda}Div(w), p)), \end{equation}$

(4.11)

with $Div(w) = \sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}}$ . □

5. Flow time bilinear control problem

Now, we are able to discuss the case of bilinear time control of the type $Q = Q(t)$ . We want to reach a flow spatial state target prescribed on the whole domain $\Theta$ at a fixed time $m$ .

In such case, the set of controls (2.1) becomes

$\begin{equation} Q \in U_{p} = \{ Q\in L^{\infty}([0,m]){\mbox{ such that }} -p\leq Q(t) \leq p {\mbox{ for }} t \in (0, m)\}, \end{equation}$

(5.1)

with $p$ as a positive constant.

The cost to minimize is

$\begin{align} C_{\lambda}(Q) = \frac{1}{2}\Big\|\nabla u(x,m)-u^{d}\Big\|^{2}_{(L^2(\Theta))^{n}} + \frac{\lambda}{2}\int_{0}^{m} Q^2(t) dt\\ = \frac{1}{2}\sum\limits_{i = 1}^{n} \Big\|\frac{\partial u}{\partial x_{i}}(x,m) - u_{i}^{d}\Big\|^{2}_{L^2(\Theta)} + \frac{\lambda}{2}\int_{0}^{m} Q^2(t)dt, \end{align}$

(5.2)

where $u^d = (u_{1}^{d}, ....u_{n}^{d})$ is the flow spatial target in $L^2(\Theta)$ . The flow control problem is

$\begin{equation} \min\limits_{Q\ \in U_{p} }C_{\lambda}(Q), \end{equation}$

(5.3)

where $C_{\lambda}$ is the flow penalizing cost defined by (5.2), and $U_{p}$ is defined by (5.1).

Corollary 5.1. The solution of the flow time control problem (5.3) is

$\begin{align} Q(t) = max(-p, min(\int_{\Theta}\frac{-u_{t}}{\lambda} (\sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}})dx,p)) \end{align}$

(5.4)

with $u$ as the solution of (2.2) perturbed by $Q(t)$ and $w_{i}$ as the solution of

$\begin{equation} \left \{ \begin{array}{lll} {\frac{{\partial^{2} } {w_{i}}}{{\partial t^{2}}} } +\Delta^{2} w_{i} = Q(t) (w_{i})_{t}, & & (0, m)\times\Theta, \\ w_{i}(x,m) = ( \frac{\partial u}{\partial x_{i}}(x,m) - u_{i}^{d}), & & \Theta, \\ (w_{i})_{t}(x,m) = 0, & & \Theta, \\ w_{i} = \frac{\partial w_{i}}{\partial \nu } = 0, & & \Gamma. \end{array} \right. \end{equation}$

(5.5)

Proof. Similar to the approach used in the proof of Theorem 4.1, we deduce that

$\begin{align} 0\leq \int_{0}^{m}d(t)[\int_\Theta u_{t}(\sum\limits_{i = 1}^{n}\frac{\partial w_{i}(x,t)}{\partial x_{i}})dx+\lambda Q ]dt, \end{align}$

(5.6)

where $d(t)\in L^{\infty}(0, m)$ , a control function such that $Q+k d \in U_{p}$ with a small positive constant $k$ . □

Remark 5.1. (1) In the case of spatiotemporal target, we remark that the error $(\frac{\partial u}{\partial x_{i}}(x, t) - u_{i}^{d})$ between the state and the desired one becomes a the change of velocity induced by the known forces acting on system (4.6).

(2) In the case of a prescribed time $m$ targets, we remark that the error $(\frac{\partial u}{\partial x_{i}}(x, m) - u_{i}^{d})$ between the state and the desired one becomes a Dirichlet boundary condition in the adjoint equation (5.5).

6. Partial flow optimal control problem

This section establishes the flow partial optimal control problem governed by the plate equation (2.2). Afterward we characterize the solution. Let $\theta\subset \Theta$ be an open subregion of $\Theta$ and we define

$\begin{array}{lll} \widetilde{P_{\theta}}: (L^2(\Theta)) & \longrightarrow & (L^2(\theta)) \\ \qquad \qquad u & \longrightarrow & \widetilde{P}_{\theta} u = u|_{\theta}, \end{array}$

and

$\begin{array}{lll} P_{\theta}: (L^2(\Theta))^{n} & \longrightarrow &(L^2(\theta))^{n} \\ \qquad \qquad u & \longrightarrow & P_{\theta} u = u|_{\theta}. \end{array}$

We define the adjoint of $P_{\theta}$ by

$P_{\theta}^{*}u = \left\{\begin{array}{lll} u \, \, in \, \, \Theta, \\ 0 \in \Theta \setminus \theta. \end{array} \right.$

Definition 6.1. The plate equation (2.2) is said to flow weakly partially controllable on $\theta \subset \Theta$ , if for $\forall \, \, \beta > 0$ , one can find an optimal control $Q\in L^{2}(0, m)$ such that

$|P_\theta \nabla u_Q(m)-u^d||_{(L^2(\theta))^{n}}\leq \beta,$

where $u^d = (z_{1}^{d}, ...., u_{n}^{d})$ is the desired flow in $(L^2(\theta))^{n}$ .

For $U_{p}$ defined by (5.1), we take the partial flow optimal control problem

$\begin{equation} \min\limits_{Q \in U_{p}} C_{\lambda}(Q), \end{equation}$

(6.1)

and the partial flow cost $C_{\lambda}$ is

$\begin{align} C_{\lambda}(Q) = \frac{1}{2}\Big\|P_\theta\nabla u(m)-u^{d}\Big\|^{2}_{(L^2(\theta))^{n}} + \frac{\lambda}{2} \int_{0}^{m} Q^2(t) dt\\ = \frac{1}{2}\sum\limits_{i = 1}^{n} \Big\|\widetilde{P}_{\theta}\frac{\partial u(T)}{\partial x_{i}} - u_{i}^{d}\Big\|^{2}_{(L^2(\theta))} + \frac{\lambda}{2} \int_{0}^{m} Q^2(t) dt. \end{align}$

(6.2)

Next, we consider the family of optimality systems

$\begin{equation} \left \{ \begin{array}{lll} {\frac{{\partial^{2} } {w_{i}} }{{\partial t^{2}}} } = \Delta^{2} w_{i}+Q(t) (w_{i})_{t},& & (0, m)\times\Theta,\\ w_{i}(x,m) = ( \frac{\partial u(m)}{\partial x_{i}} - \widetilde{P}_{\theta}^{*}u_{i}^{d}), & & \Theta, \\ (w_{i})_{t}(x,m) = 0, & & \Theta, \\ w_{i}(x,t) = \frac{\partial w_{i}(x,t)}{\partial \nu } = 0, & & \Gamma. \end{array} \right. \end{equation}$

(6.3)

Lemma 6.1. Let the cost $C_{\lambda}(Q)$ defined by (6.2) and the control $Q(t) \in U_{p}$ be the solution of (6.1). We can write

$\begin{align} \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q )}{k} = \sum\limits_{i = 1}^{n} \int_\theta \widetilde{P}^*_{\theta}\widetilde{P}_{\theta}\left[ \int_{0}^{m}\frac{\partial^{2} w_{i}}{\partial t^{2}} \frac{\partial v(x,t)}{\partial x_{i}} dt+ \int_{0}^{m}w_{i} \frac{\partial }{\partial x_{i}}( \frac{\partial^{2} v}{\partial t^{2}}) dt\right]dx\\ + \int_{0}^{m}\lambda d Q dt, \end{align}$

(6.4)

where the solution of (4.1) is $v$ , and the solution of (6.3) is $w_{i}$ .

Proof. The functional $C_{\lambda}(Q)$ given by (6.2), is of the form:

$\begin{align} C_{\lambda}(Q) = \frac{1}{2} \sum\limits_{i = 1}^{n}\int_{\theta}(\widetilde{P}_{\theta}\frac{\partial u}{\partial x_{i}} - u_{i}^{d} )^2 dx + \frac{\lambda}{2} \int^{m}_{0} Q^2(t) dt. \end{align}$

(6.5)

Choose $u_{k} = u(Q+k d)$ and $u = u(Q)$ . By (6.5), we deduce

$\begin{align} \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} = \lim\limits_{k\longrightarrow 0}\sum\limits_{i = 1}^{n} \frac{1}{2}\int_{\theta}\frac{(\widetilde{P}_\theta \frac{\partial u_{k}}{\partial x_{i}}-u_i^d)^{2}-(\widetilde{P}_\theta \frac{\partial u}{\partial x_{i}}-u_i^d)^{2}}{k}dx\\ + \lim\limits_{k\longrightarrow 0}\frac{\lambda}{2}\int_{0}^{m}\frac{(Q+k d)^2-Q^2}{k}dt. \end{align}$

(6.6)

Furthermore,

$\begin{align} \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} \\ = \lim\limits_{k\longrightarrow 0}\sum\limits_{i = 1}^{n}\frac{1}{2}\int_{\theta}\widetilde{P}_{\theta}\frac{( \frac{\partial u_{k}}{\partial x_{i}}- \frac{\partial u}{\partial x_{i}})}{k}(\widetilde{P}_{\theta} \frac{\partial u_{k} }{\partial x_{i}}+ \widetilde{P}_{\theta} \frac{\partial u}{\partial x_{i}}-2u_i^d)dx \\ + \lim\limits_{k\longrightarrow 0}\frac{1}{2}\int_{0}^{m}(2\lambda d Q+k \lambda d^2)dt\\ = \sum\limits_{i = 1}^{n}\int_\theta \widetilde{P}_{\theta} \frac{\partial v(x,m)}{\partial x_{i}} \widetilde{P}_{\theta}( \frac{\partial u(x,m)}{\partial x_{i}}-\widetilde{P}_{\theta}^{*}u_i^d)dx+\int_{0}^{m}\lambda dQ dt \\ = \sum\limits_{i = 1}^{n} \int_\theta \widetilde{P}_{\theta} \frac{\partial v(x,m)}{\partial x_{i}} \widetilde{P}_{\theta}w_{i}(x,m)dx+\lambda \int_{0}^{m}dQ dt. \end{align}$

(6.7)

Using (6.3) to (6.7), we conclude

$\begin{align} \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k} = \sum\limits_{i = 1}^{n} \int_\theta \widetilde{P}^*_{\theta}\widetilde{P}_{\theta}\left[ \int_{0}^{m}\frac{\partial^{2} w_{i}}{\partial t^{2}} \frac{\partial v(x,t)}{\partial x_{i}} dt+ \int_{0}^{m}w_{i} \frac{\partial }{\partial x_{i}}( \frac{\partial^{2} v}{\partial t^{2}}) dt\right]dx\\\\ + \int_{0}^{m}\lambda d Qdt.\\ \end{align}$

(6.8)

□

Theorem 6.1. Consider the set $U_{p}$ , of partial admissible control defined as (5.1) and $u = u (Q)$ is its associate solution of (2.2), then the solution of (6.1) is

$\begin{equation} Q_{\varepsilon}(t) = max(-p,min( \frac{-1}{\lambda}\widetilde{P}_{\theta}(u_{t})(\widetilde{P}_{\theta} Div(w)),p)), \end{equation}$

(6.9)

where $Div(w) = \sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}}$ .

Proof. Let $d \in U_{p}$ and $Q + k d \in U_{p}$ for $k > 0$ . The cost $C_{\lambda}$ at its minimum $Q$ , verifies

$\begin{equation} 0 \leq \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k}. \end{equation}$

(6.10)

From Lemma 6.1, substituting $\frac{\partial^{2} v}{\partial t^{2}}$ , by its value in system (4.1), we deduce that

$\begin{align} 0 \leq \lim\limits_{k\longrightarrow 0} \frac{C_{\lambda}(Q+k d)-C_{\lambda}(Q)}{k}\\ = \sum\limits_{i = 1}^{n}\int_\theta \widetilde{P}^*_{\theta}\widetilde{P}_{\theta}\left[\int_{0}^{m} \frac{\partial v}{\partial x_{i}}\frac{\partial^{2} w_{i} }{\partial t^{2}}dt+ \int_{0}^{m}(-\Delta^{2} \frac{\partial v}{\partial x_{i}}+Q(t) \frac{\partial }{\partial x_{i}}(v_{t}) + d(t)\frac{\partial }{ \partial x_{i}}(u_{t}))w_{i} dt\right]dx\\ + \int_{0}^{m}\lambda d Q dt, \end{align}$

(6.11)

and system (6.3) gives

$\begin{align} 0 \leq \sum\limits_{i = 1}^{n} \int_\theta \widetilde{P}^*_{\theta}\widetilde{P}_{\theta}\left[\int_{0}^{m} \frac{\partial v}{\partial x_{i}}(\frac{\partial^{2} w_{i} }{\partial t^{2}}-\Delta^{2} w_{i}-Q(t) (w_{i})_{t})dt+d(t) \frac{\partial }{\partial x_{i}}(u_{t})w_{i} dt\right]dx+\int_{0}^{m}\lambda d Q dt.\\ \\ = \sum\limits_{i = 1}^{n} \int_\theta \widetilde{P}^*_{\theta}\widetilde{P}_{\theta} \int_{0}^{m} (h(t)u_{t})\frac{\partial w_{i}}{\partial x_{i}} dt+ \int_{0}^{m}\lambda d Q dt.\\ \\ = \int_{0}^{m} h(t)\int_\theta\left[ u_{t}\widetilde{P}^*_{\theta}\widetilde{P}_{\theta} \sum\limits_{i = 1}^{n}\frac{\partial w_{i}}{\partial x_{i}} + \lambda Q\right] dx dt, \end{align}$

(6.12)

which gives the optimal control

$\begin{equation} Q_{\varepsilon}(t) = max(-p,min( \frac{-1}{\lambda}\widetilde{P}_{\theta}(u_{t})(\widetilde{P}_{\theta} Div(w)),p)). \end{equation}$

(6.13)

□

7. Conclusions

This paper studied the optimal control problem governed by an infinite dimensional bilinear plate equation. The objective was to command the flow state of the bilinear plate equation to the desired flow using different types of bounded feedback. The problem flow optimal control governed by a bilinear plate equation was considered and solved in two cases using the adjoint method. The first case considered a spatiotemporal control function and looked to reach a flow target on the whole domain. The second case considered a time control function and looks to reach a prescribed target at a fixed final time. As an application, the partial flow control problem was established and solved using the proposed method. More applications can be examined, for example. the case of fractional hyperbolic systems.

Use of AI tools declaration

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

Acknowledgments

This work was funded by the Deanship of Scientific Research at Jouf University under Grant Number (DSR2022-RG-0119).

Conflict of interest

The authors affirm that they have no conflicts of interest to disclose.

References

[1]	M. Aringer, S. R. Johnson, Classifying and diagnosing systemic lupus erythematosus in the 21st century, Rheumatology, 59 (2020), v4–v11. http://doi.org/10.1093/rheumatology/keaa379 doi: 10.1093/rheumatology/keaa379
[2]	G. Ruiz-Irastorza, G. Bertsias, Treating systemic lupus erythematosus in the 21st century: new drugs and new perspectives on old drugs, Rheumatology, 59 (2020), v69–v81. http://doi.org/10.1093/rheumatology/keaa403 doi: 10.1093/rheumatology/keaa403
[3]	P. M. van der Meulen, A. M. Barendregt, E. Cuadrado, C. Magro-Checa, G. M. Steup-Beekman, D. Schonenberg-Meinema, et al., Protein array autoantibody profiles to determine diagnostic markers for neuropsychiatric systemic lupus erythematosus, Rheumatology, 56 (2017), 1407–1416. http://doi.org/10.1093/rheumatology/kex073 doi: 10.1093/rheumatology/kex073
[4]	M. E. Kathleen, A. Janice, H. Margaret, B. Jane, L. Peter, R. Anisur, et al., Flares in patients with systemic lupus erythematosus, Rheumatology, 60 (2021), 3262–3267. http://doi.org/10.1093/rheumatology/keaa777 doi: 10.1093/rheumatology/keaa777
[5]	A. Kernder, E. Elefante, G. Chehab, C. Tani, M. Mosca, M. Schneider, The patient's perspective: are quality of life and disease burden a possible treatment target in systemic lupus erythematosus?, Rheumatology, 59 (2020), v63–v68. http://doi.org/10.1093/rheumatology/keaa427 doi: 10.1093/rheumatology/keaa427
[6]	M. Bruschi, G. Moroni, R. A. Sinico, F. Franceschini, M. Fredi, A. Vaglio, et al., Serum igg2 antibody multi-composition in systemic lupus erythematosus and in lupus nephritis (part 2): prospective study, Rheumatology, 60 (2021), 3388–3397. http://doi.org/10.1093/rheumatology/keaa793 doi: 10.1093/rheumatology/keaa793
[7]	L. Arnaud, M. G. Tektonidou, Long-term outcomes in systemic lupus erythematosus: trends over time and major contributors, Rheumatology, 59 (2020), 29–38. http://doi.org/10.1093/rheumatology/keaa382 doi: 10.1093/rheumatology/keaa382
[8]	N. Sarbu, F. Alobeidi, P. Toledano, G. Espinosa, I. Giles, A. Rahman, et al., Brain abnormalities in newly diagnosed neuropsychiatric lupus: systematic mri approach and correlation with clinical and laboratory data in a large multicenter cohort, Autoimmun. Rev., 14 (2015), 153–159. http://doi.org/10.1016/j.autrev.2014.11.001 doi: 10.1016/j.autrev.2014.11.001
[9]	J. A. Mikdashi, Altered functional neuronal activity in neuropsychiatric lupus: a systematic review of the fmri investigations, Semin. Arthritis Rheum., 45 (2016), 455–462. http://doi.org/10.1016/j.semarthrit.2015.08.002 doi: 10.1016/j.semarthrit.2015.08.002
[10]	M. Govoni, J. G. Hanly, The management of neuropsychiatric lupus in the 21st century: still so many unmet needs?, Rheumatology, 59 (2020), v52–v62. http://doi.org/10.1093/rheumatology/keaa404 doi: 10.1093/rheumatology/keaa404
[11]	M. H. Liang, M. Corzillius, S. C. Bae, R. A. Lew, P. R. Fortin, C. Gordon, et al., The american college of rheumatology nomenclature and case definitions for neuropsychiatric lupus syndromes, Arthritis Rheum., 42 (1999), 599–608. http://doi.org/10.1002/1529-0131(199904)42:4 < 599::AID-ANR2 > 3.0.CO; 2-F doi: 10.1002/1529-0131(199904)42:4 < 599::AID-ANR2 > 3.0.CO; 2-F
[12]	E. Moore, M. W. Huang, C. Putterman, Advances in the diagnosis, pathogenesis and treatment of neuropsychiatric systemic lupus erythematosus, Curr. Opin. Rheumatol., 32 (2020), 152–158. http://doi.org/10.1097/BOR.0000000000000682 doi: 10.1097/BOR.0000000000000682
[13]	H. Jeltsch-David, S. Muller, Neuropsychiatric systemic lupus erythematosus: pathogenesis and biomarkers, Nat. Rev. Neurol., 10 (2014), 579–596. http://doi.org/10.1038/nrneurol.2014.148 doi: 10.1038/nrneurol.2014.148
[14]	C. Magro-Checa, E. J. Zirkzee, L. J. Beaart-van de Voorde, H. A. Middelkoop, N. J. van der Wee, M. V. Huisman, et al., Value of multidisciplinary reassessment in attribution of neuropsychiatric events to systemic lupus erythematosus: prospective data from the leiden npsle cohort, Rheumatology, 56 (2017), 1676–1683. http://doi.org/10.1093/rheumatology/kex019 doi: 10.1093/rheumatology/kex019
[15]	A. N. Culshaw, D. B. Roychowdhury, P80 value of the clinical nurse specialist role in the care of patients with systemic lupus erythematosus: the patient experience, Rheumatology, 59 (2020), keaa111.078. http://doi.org/10.1093/rheumatology/keaa111.078 doi: 10.1093/rheumatology/keaa111.078
[16]	Y. Cheng, A. Cheng, Y. Jia, L. Yang, Y. Ning, L. Xu, et al., ph-responsive multifunctional theranostic rapamycin-loaded nanoparticles for imaging and treatment of acute ischemic stroke, ACS Appl. Mater. Interfaces, 13 (2021), 56909–56922. http://doi.org/10.1021/acsami.1c16530 doi: 10.1021/acsami.1c16530
[17]	J. Luyendijk, S. Steens, W. Ouwendijk, G. Steup-Beekman, E. Bollen, J. Van Der Grond, et al., Neuropsychiatric systemic lupus erythematosus: lessons learned from magnetic resonance imaging, Arthritis Rheum., 63 (2011), 722–732. http://doi.org/10.1002/art.30157 doi: 10.1002/art.30157
[18]	H. Lu, Z. Ge, Y. Song, D. Jiang, T. Zhou, J. Qin, A temporal-aware lstm enhanced by loss-switch mechanism for traffic flow forecasting, Neurocomputing, 427 (2021), 169–178. http://doi.org/10.1016/j.neucom.2020.11.026 doi: 10.1016/j.neucom.2020.11.026
[19]	H. Lu, D. Huang, S. Youyi, D. Jiang, T. Zhou, J. Qin, St-trafficnet: A spatial-temporal deep learning network for traffic forecasting, Electronics, 9 (2020), 1–17. http://doi.org/10.3390/electronics9091474 doi: 10.3390/electronics9091474
[20]	Y. Song, Z. Yu, T. Zhou, J. Y. C. Teoh, B. Lei, C. Kup-Sze, et al., Cnn in ct image segmentation: Beyond loss function for exploiting ground truth images, in 2020 IEEE International Symposium on Biomedical Imaging (ISBI), (2020), 325–328. http://doi.org/10.1109/ISBI45749.2020.9098488
[21]	T. Zhou, G. Han, B. N. Li, Z. Lin, E. J. Ciaccio, P. H. Green, et al., Quantitative analysis of patients with celiac disease by video capsule endoscopy: A deep learning method, Comput. Biol. Med., 85 (2017), 1–6. http://doi.org/10.1016/j.compbiomed.2017.03.031 doi: 10.1016/j.compbiomed.2017.03.031
[22]	G. Tu, J. Wen, H. Liu, S. Chen, L. Zheng, D. Jiang, Exploration meets exploitation: Multitask learning for emotion recognition based on discrete and dimensional models, Knowl. Based Syst., 235 (2022), 107598. http://doi.org/10.1016/j.knosys.2021.107598 doi: 10.1016/j.knosys.2021.107598
[23]	W. Fang, W. Zhuo, J. Yan, Y. Song, D. Jiang, T. Zhou, Attention meets long short-term memory: A deep learning network for traffic flow forecasting, Phys. A, 587 (2022), 126485. http://doi.org/10.1016/j.physa.2021.126485 doi: 10.1016/j.physa.2021.126485
[24]	Y. LeCun, Y. Bengio, G. Hinton, Deep learning, Nature, 521 (2015), 436–444. http://doi.org/10.1038/nature14539
[25]	C. Cortes, V. Vapnik, Support-vector networks, Mach. Learn., 20 (1995), 273–297. http://doi.org/10.1007/BF00994018
[26]	L. Zhu, P. Spachos, Support vector machine and yolo for a mobile food grading system, Internet Things, 13 (2021), 100359. http://doi.org/10.1016/j.iot.2021.100359 doi: 10.1016/j.iot.2021.100359
[27]	W. Cai, D. Yu, Z. Wu, X. Du, T. Zhou, A hybrid ensemble learning framework for basketball outcomes prediction, Phys. A, 528 (2019), 1–8. http://doi.org/10.1016/j.physa.2019.121461 doi: 10.1016/j.physa.2019.121461
[28]	G. Tan, S. Zheng, B. Huang, Z. Cui, H. Dou, X. Yang, et al., Hybrid ga-svr: An effective way to predict short-term traffic flow, in 21st International Conference on Algorithms and Architectures for Parallel Processing (ICA3PP 2021), (2021), 1–11.
[29]	G. N. Kouziokas, Svm kernel based on particle swarm optimized vector and bayesian optimized svm in atmospheric particulate matter forecasting, Appl. Soft Comput., 93 (2020), 106410. http://doi.org/10.1016/j.asoc.2020.106410 doi: 10.1016/j.asoc.2020.106410
[30]	W. Cai, J. Yang, Y. Yu, Y. Song, T. Zhou, J. Qin, Pso-elm: A hybrid learning model for short-term traffic flow forecasting, IEEE Access, 8 (2020), 6505–6514. http://doi.org/10.1109/ACCESS.2019.2963784 doi: 10.1109/ACCESS.2019.2963784
[31]	H. Faris, M. A. Hassonah, A. Z. Ala'M, S. Mirjalili, I. Aljarah, A multi-verse optimizer approach for feature selection and optimizing svm parameters based on a robust system architecture, Neural Comput. Appl., 30 (2018), 2355–2369. http://doi.org/10.1007/s00521-016-2818-2 doi: 10.1007/s00521-016-2818-2
[32]	B. K. Petersen, J. Yang, W. S. Grathwohl, C. Cockrell, C. Santiago, G. An, et al., Precision medicine as a control problem: Using simulation and deep reinforcement learning to discover adaptive, personalized multi-cytokine therapy for sepsis, preprint, arXiv: 1802.10440.
[33]	C. Yu, Y. Dong, J. Liu, G. Ren, Incorporating causal factors into reinforcement learning for dynamic treatment regimes in hiv, BMC Med. Inf. Decis. Making, 19 (2019), 19–29. http://doi.org/10.1186/s12911-019-0755-6 doi: 10.1186/s12911-019-0755-6
[34]	G. Maicas, G. Carneiro, A. P. Bradley, J. C. Nascimento, I. Reid, Deep reinforcement learning for active breast lesion detection from dce-mri, in International conference on medical image computing and computer-assisted intervention, (2017), 665–673.
[35]	H. Dou, J. Ji, H. Wei, F. Wang, J. Wang, T. Zhou, Transfer inhibitory potency prediction to binary classification: A model only needs a small training set, Comput. Methods Programs Biomed..
[36]	K. Liu, Y. Fu, P. Wang, L. Wu, R. Bo, X. Li, Automating feature subspace exploration via multi-agent reinforcement learning, in Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, (2019), 207–215. http://doi.org/10.1145/3292500.3330868
[37]	Z. Tao, L. Huiling, W. Wenwen, Y. Xia, Ga-svm based feature selection and parameter optimization in hospitalization expense modeling, Appl. Soft Comput., 75 (2019), 323–332. http://doi.org/10.1016/j.asoc.2018.11.001 doi: 10.1016/j.asoc.2018.11.001
[38]	C. Sukawattanavijit, J. Chen, H. Zhang, Ga-svm algorithm for improving land-cover classification using sar and optical remote sensing data, IEEE Geosci. Remote Sens. Lett., 14 (2017), 284–288. http://doi.org/10.1109/LGRS.2016.2628406 doi: 10.1109/LGRS.2016.2628406
[39]	C. Meng, Y. Hu, Y. Zhang, F. Guo, Psbp-svm: A machine learning-based computational identifier for predicting polystyrene binding peptides, Frontiers in bioengineering and biotechnology, 8 (2020), 245. http://doi.org/10.3389/fbioe.2020.00245 doi: 10.3389/fbioe.2020.00245
[40]	L. Cai, Q. Chen, W. Cai, X. Xu, T. Zhou, J. Qin, Svrgsa: a hybrid learning based model for short-term traffic flow forecasting, IET Intell. Transp. Syst., 13 (2019), 1348–1355. http://doi.org/10.1049/iet-its.2018.5315 doi: 10.1049/iet-its.2018.5315
[41]	S. Katoch, S. S. Chauhan, V. Kumar, A review on genetic algorithm: past, present, and future, Multimedia Tools Appl., 80 (2021), 8091–8126. http://doi.org/10.1007/s11042-020-10139-6 doi: 10.1007/s11042-020-10139-6
[42]	M. A. Ehyaei, A. Ahmadi, M. A. Rosen, A. Davarpanah, Thermodynamic optimization of a geothermal power plant with a genetic algorithm in two stages, Processes, 8 (2020), 1277. http://doi.org/10.3390/pr8101277 doi: 10.3390/pr8101277
[43]	R. S. Sutton, A. G. Barto, Reinforcement learning: An introduction, MIT press, (2018), http://doi.org/10.1109/TNN.1998.712192
[44]	V. François-Lavet, P. Henderson, R. Islam, M. G. Bellemare, J. Pineau, An introduction to deep reinforcement learning, Found. Trends Mach. Learn., 11 (2018), 219–354. http://doi.org/10.1561/2200000071 doi: 10.1561/2200000071
[45]	S. Gronauer, K. Diepold, Multi-agent deep reinforcement learning: a survey, Artif. Intell. Rev., 1–49. http://doi.org/10.1007/s10462-021-09996-w
[46]	Z. Yin, J. Hou, Recent advances on svm based fault diagnosis and process monitoring in complicated industrial processes, Neurocomputing, 174 (2016), 643–650. http://doi.org/10.1016/j.neucom.2015.09.081 doi: 10.1016/j.neucom.2015.09.081
[47]	Z. Liu, L. Wang, Y. Zhang, C. P. Chen, A svm controller for the stable walking of biped robots based on small sample sizes, Appl. Soft Comput., 38 (2016), 738–753. http://doi.org/10.1016/j.asoc.2015.10.029 doi: 10.1016/j.asoc.2015.10.029
[48]	S. M. Erfani, S. Rajasegarar, S. Karunasekera, C. Leckie, High-dimensional and large-scale anomaly detection using a linear one-class svm with deep learning, Pattern Recognit., 58 (2016), 121–134. http://doi.org/10.1016/j.patcog.2016.03.028 doi: 10.1016/j.patcog.2016.03.028
[49]	A. Coronato, A. Cuzzocrea, An innovative risk assessment methodology for medical information systems, IEEE Trans. Knowl. Data Eng., (2020). http://doi.org/10.1109/TKDE.2020.3023553
[50]	Z. Zhuo, L. Su, Y. Duan, J. Huang, X. Qiu, S. Haller, et al., Different patterns of cerebral perfusion in sle patients with and without neuropsychiatric manifestations, Hum. brain Mapp., 41 (2020), 755–766. http://doi.org/10.1002/hbm.24837 doi: 10.1002/hbm.24837
[51]	E. Kozora, M. C. Ellison, S. West, Depression, fatigue, and pain in systemic lupus erythematosus (sle): relationship to the american college of rheumatology sle neuropsychological battery, Arthritis Rheum., 55 (2006), 628–635. http://doi.org/10.1002/art.22101 doi: 10.1002/art.22101
[52]	A. Anaby-Tavor, B. Carmeli, E. Goldbraich, A. Kantor, G. Kour, S. Shlomov, et al., Do not have enough data? deep learning to the rescue!, in Proceedings of the AAAI Conference on Artificial Intelligence, 34 (2020), 7383–7390. http://doi.org/10.1609/aaai.v34i05.6233

This article has been cited by:

Fatima Hussain, Suha Shihab, Transforming Controlled Duffing Oscillator to Optimization Schemes Using New Symmetry-Shifted G(t)-Polynomials, 2024, 16, 2073-8994, 915, 10.3390/sym16070915

Reader Comments

Your name:*

Email:*
© 2022 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)