A PSO-optimized novel PID neural network model for temperature control of jacketed CSTR: design, simulation, and a comparative study

Abstract

This paper proposes a Particle Swarm Optimization (PSO) tuned novel Proportional Integral Derivative (PID) like neural network (PID-NN). The structure of proposed PID-NN is very simple having only 3 neurons in the hidden layer and a single output neuron. The proportional, integral, and derivative gains of the PID controller are represented by the three weights in the neural network's output layer, respectively. The suggested approach uses the PSO method to optimize the output layer weights, which correspond to PID gains. The non-linear Continuous Stirred Tank Reactor (CSTR) plant, one of the most popular chemical industry processes, is utilized to test the suggested approach. A jacketed CSTR's temperature is controlled via a Particle Swarm Optimization tuned PID like neural network (PSO-NN-PID) controller. In terms of time domain specifications, the performance of the PSO-based NN-PID controller, the back propagation-based NN-PID controller (BP-NN-PID), and the conventionally tuned PID controller are compared. Mean square error is the objective function used in PSO-NN-PID and BP-NN-PID to optimize PID settings. The results show that the overshoot decreases from 44.13% in case of Zeigler-Nichols tuned PID controller to 26.33% in case of BPNN-PID controller, and further reduces to 23.13% in case of proposed PSO-based NN-PID controller. The decrease in rise time is observed from 0.2727 seconds in case of BPNN tuned PID to 0.1283 seconds in case of proposed PSO-NN-PID controller.

A Dual-Competition-Based Particle Swarm Optimizer for Large-Scale Optimization

“…Particle swarm optimization (PSO) has been widely applied in engineering optimization in past decades due to their simplicity and efficiency [1][2][3][4][5]. On the one hand, PSO shows better robustness and computational efficiency in comparison to gradient-based algorithms [6].…”