Abstract:It is necessary to consider the robustness in the tracking problem, which can effectively suppress the external disturbance to ensure the tracking performance. Different from previous tracking control methods, considering the robustness, completely unknown nonlinear system dynamics and constrained controller, we propose a data-based echo state network (ESN) approximated algorithm for a class of robust tracking problems. First, the robust tracking control problem (RTCP) is transformed into the optimal control problem of the according nominal system by designing a elaborate value function. To obtain the optimal control policy, we have to solve a Hamilton-Jacobi-Bellman equation (HJBE) about the augmented nominal system. It is well-known that modelling the accurate dynamics for the practical engineering applications is usually difficult, so the model-free integral reinforcement learning (IRL) algorithm is used to learn the optimal control policy and performance function simultaneously by only using systems data. In this IRL algorithm, a reservoir computing based ESN is used to approximate the performance function and control input. Contrast to other neural networks, ESN need not consider the choice of activation function, which can greatly reduce the difficulty and effort of neural network structure design. The output weights of the ESNs are iteratively updated towards the optimal ones by using least square algorithm and the pre-collected off-line system data. Then, using the converged output weights and ESNs, the tracking control input can be derived without knowing any system dynamic information. Finally, we demonstrate that the given system can be controlled to track the desired trajectory well under the proposed method by using two simulation examples.

Regularized Taylor Echo State Networks for Predictive Control of Partially Observed Systems.

Echo State Network-Based Robust Tracking Control for Unknown Constrained Nonlinear Systems by Using Integral Reinforcement Learning

Long-Short Term Echo State Network for Time Series Prediction

Prediction for noisy nonlinear time series by echo state network based on dual estimation

Physics-Constrained Taylor Neural Networks for Learning and Control of Dynamical Systems

Simple Recurrent Neural Network‐Based Adaptive Predictive Control for Nonlinear Systems

Physics-Informed Echo State Networks for Modeling Controllable Dynamical Systems

Multi-dimensional Taylor Network Adaptive Predictive Control for Single-Input Single-Output Nonlinear Systems with Input Time-Delay.

T-distributed stochastic neighbor embedding echo state network with state matrix dimensionality reduction for time series prediction

L1/2 Norm Regularized Echo State Network for Chaotic Time Series Prediction.

Neuroadaptive Tracking Control of Affine Nonlinear Systems Using Echo State Networks Embedded With Multiclustered Structure and Intrinsic Plasticity

Physics-Informed LSTM Network-Based Nonlinear Model Predictive Control

Hierarchical Echo State Network with Sparse Learning: A Method for Multidimensional Chaotic Time Series Prediction.

Echo State Networks Based Data-Driven Adaptive Fault Tolerant Control with Its Application to Electromechanical System

Adaptive Echo State Network Control for a Class of Pure-Feedback Systems with Input and Output Constraints.

Chaotic Time Series Prediction Using a Novel Echo State Network Model with Input Reconstruction, Bayesian Ridge Regression and Independent Component Analysis

Variability of echo state network prediction horizon for partially observed dynamical systems

Adaptive NN control for a class of strict-feedback nonlinear systems with time delays

Fault-tolerant Adaptive Tracking Control of Euler-Lagrange Systems - an Echo State Network Approach Driven by Reinforcement Learning

Chaotic Time Series Prediction Based on a Novel Robust Echo State Network

Recursive D-Step-ahead Predictive Control of MIMO Nonlinear Systems with Input Time-Delay Via Multi-Dimensional Taylor Network Extended from PID