Packet-dropouts compensation for networked control system via deep ReLU neural network

Abstract

This paper introduces a packet-dropouts compensation strategy for networked control systems. To achieve robustness with respect to packet-dropouts, the predictions of the feedback losses in network transmission are included in the data packets. To achieve high-precision predictions, a deep ReLU neural network is used to build the relationships between the system input and feedback losses. We show how to design the parameters of the deep neural network to ensure stability of the resulting feedback control systems when the number of packet-dropouts is bounded. Simulation results indicate that the proposed compensation strategy can achieve much better control performances than the widely used zero-input or hold-input strategies, especially when the system inputs include abundant noises.