Improved NN-PID control of MIMO systems with PSO-based initialisation of weights

摘要

To train the neural networks (NNs) standard back propagation (BP) algorithm and its variations are widely used where initial weights are generated as random in nature The convergence of these algorithms is very sensitive to the initial weights. In this paper, particle swarm optimisation (PSO) algorithm has been used to initialise the weights by optimisation. Various available algorithms and the proposed one have been tested and compared for the implementation of NN-PID control of two discrete-time non-linear coupled MIMO systems. Simulation results show that the controlled performance of BP algorithm for non-linear, coupled MIMO systems can be significantly improved, with the use of PSO algorithm to initialise the weights.