Echo-state-network-based iterative learning control of distributed systems

Abstract

The paper proposes an effective modeling and control procedure for the distributed-parameter systems using the echo-state network. The main idea is to reconstruct the spatiotemporal dynamics defined in a given multi-dimensional domain. In the investigated problem positions of both sensors and actuators are fixed allowing to delegate the complex system dynamics to echo-state network. Imposing a proper partitioning of the spatial domain, a specific topology of a neural network is used to form a reservoir capable to follow not only temporal but also spatial dynamics of the system. Based on available historical data, neural network model is initially trained and then used to derive the control law in the framework of iterative learning control. The echo-state network can be retrained after a particular control iterate in order to reduce model uncertainty and to fit it to the current operating conditions as much as possible. The performance of the proposed approach is tested and evaluated on the example of the squared clamped plate control.