Neural-Based Predefined-Time Distributed Optimization of High-Order Nonlinear Multiagent Systems

Abstract

This paper addresses a predefined-time distributed optimization problem for high-order nonlinear multiagent systems (MASs). First, by means of a distributed proportional integration (PI) protocol, a reference model is constructed to evaluate the global optimal solution for MASs. Then, the resulting measurement is fed into a prefilter to produce a reconstructed optimal reference signal and its high-order derivatives. Instead of designing the updated law with σ -modification to deal with unknown nonlinearities, a gradient descent algorithm is developed to train the weights of neural networks (NNs) to achieve higher function approximation accuracy. Moreover, in the framework of prefiltering, an NN-based predefined-time control strategy is built using the backstepping technique to guarantee that all agents' outputs can reach optimal consensus in predefined time. Finally, simulation examples validate the effectiveness of the presented approach.