Dynamic Learning From Adaptive Neural Control for Discrete-Time Strict-Feedback Systems

Huang

Zhao

et al. 2021

Self Cite

In this paper, an event-triggered adaptive neural output feedback control scheme is developed for a class of uncertain discrete-time strict-feedback nonlinear systems subject to immeasurable system states and network resource limitation. An i-step ahead predictor is synthesized to obtain the future signal of the reference orbit. By combining the neural observer and the variable substitution technology, an event-triggered adaptive neural control scheme is developed, thereby estimating the immeasurable system states and avoiding the n-step delays of the existing controller. To promote the transient system performance, an improved triggering condition is designed to increase the number of triggering events in the transient process. The stability analysis of the closed-loop system is divided into two parts to deal with the challenge from the simultaneous presence of state estimations, unknown system dynamics, and aperiodical controller weight updating laws. The proposed scheme achieves the state estimation, guarantees the output tracking performance with the improved transient control performance, and reduces the communication resource. Simulation studies on a numerical example and a networked robot manipulator are, respectively, implemented to show the validity of the proposed scheme. K E Y W O R D Sadaptive neural control, discrete-time nonlinear systems, event-triggered control, mismatched nonlinear uncertainties, neural observer

Section: Discussionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 97%

Observer‐based adaptive neural output‐feedback event‐triggered control for discrete‐time nonlinear systems using variable substitution

Huang

Zhao

et al. 2021

Self Cite

“…17 This theory has successfully proven that radial basis function neural networks (RBF NNs) are capable of satisfying the PE conditions along the recurrent trajectories and the neural weights can converge to the ideal values, thereby guaranteeing the learning ability of NNs. More recently, the deterministic learning theory was not only extended to some more general nonlinear systems including strict-feedback and pure-feedback, 18,19 but also employed in some practical systems including spacecraft and marine surface vessels. 20 Noting that most existing adaptive and learning control strategies do not investigate the transient and steady-state tracking performance of the considered system, which is limited in some applications since many tracking performance indicators including convergence rate, overshoot, and steady-state error are required to meet specified constraints in some engineering systems.…”

Section: Introductionmentioning

confidence: 99%

Dynamic learning‐based event‐triggered control of strict‐feedback systems with finite‐time prescribed performance

Yang

2022

Self Cite

This paper is concerned with the event-triggered neural learning control for strict-feedback nonlinear systems (SFNSs) subject to predefined tracking performance. A system transformation approach is utilized to transform the original SFNSs into the normal form, which makes it easily to verify the convergence of neural weights since only one neural network (NN) approximator is employed. Then, a novel finite-time performance function is first proposed to characterize the performance constraints of tracking error, thus guaranteeing the prescribed tracking performance. On this basis, a new lemma is given, which is crucial to ensure the stability of the closed-loop system. By using the error transformation technique, the constrained tracking control problem is converted into the stabilization problem of unconstrained error system. Subsequently, by introducing the first-order sliding mode differentiator into the backstepping procedure and with the help of the high-gain observer, a novel adaptive neural tracking control scheme is presented to ensure that the prescribed tracking performance is satisfied and all closed-loop signals are ultimately bounded. After ensuring the satisfaction of the partial persistent excitation condition for radial basis function NN, the neural weight estimates are verified to converge to the ideal values, and then the convergent weights are stored as the constant values. By utilizing the stored knowledge, an event-triggered neural learning controller is constructed to accomplish the same or similar control tasks, which can effectively improve the transient control performance, save the communication resource, and relieve the online computation at the same time. Finally, the effectiveness of the presented scheme is testified by numerical and practical examples.

“…Using the deterministic learning theory, the neural learning control (NLC) method has been extended to some continuous‐time nonlinear systems such as SFSs, 24‐26 pure‐feedback systems 27 and has also been expanded to many practical systems 28‐30 . For discrete‐time SFSs, to avoid the non‐causality problem, a NLC scheme is constructed by using the

n

‐step predictor in work, 31 and the convergence of neural weights is verified by an extended exponential stability corollary of a class of linear time‐varying systems with delays. By combining the

n

‐step input‐output predictor and the implicit function theorem, the NLC problem has also been discussed in work 32 for discrete‐time pure‐feedback systems.…”

Section: Introductionmentioning

confidence: 99%

“…By combining the

n

‐step input‐output predictor and the implicit function theorem, the NLC problem has also been discussed in work 32 for discrete‐time pure‐feedback systems. However, the result in works 31,32 shows that the neural weights based on the

n

‐step delay neural update law will converge to

n

different values, which brings many challenges to the construction of the neural learning controller. For example, the neural learning controller needs to switch constant neural weights according to the time sequence, which will cause system state chattering and may crash the control system.…”

Section: Introductionmentioning

confidence: 99%

Model‐based event‐triggered neural learning control for discrete‐time strict‐feedback systems

Shi

2022

Self Cite

In this article, the model-based event-triggered neural learning control problem is investigated for a class of discrete-time strict-feedback systems. First, with the case of sufficient network resources, an adaptive neural controller is designed with a new neural weight update law, which avoids the n-step delay of traditional adaptive neural control methods. The proposed controller performs the tracking control task while ensuring that the radial basis function neural network can accurately identify the unknown system nonlinear dynamics and the estimated neural weights can converge to their ideal values. Second, with the case of limited network resources, the convergent neural weights can be reused to construct the neural network model, event-trigger conditions, and the neural learning controller, which can improve the tracking control performance and save the communication resources. Especially, compared with the traditional event-triggered adaptive neural controller, the neural learning controller can reduce the burden of online calculations since the constant neural weights are used to construct the neural network model and trigger conditions. For the proposed control scheme, simulation results are given to verify its effectiveness.