Modified Population Extremal Optimization Based PID Neural Network for Islanded Microgrid System

This research work proposes the use of a proportional-integral-derivative (PID) like neural network for adaptive voltage control of uncertain single phase microgrid system. Gradient descent algorithm has been applied to update the weights online that explains the adaptive behavior of the controller. As gradient descent is prone to fall in local minima, a modified population extremal optimization has been applied to find the optimum initial weights for the PID neural network controller. The robustness of the controller has been examined by assigning asynchronous machine load, nonlinear load, harmonic load and unknown load. The parameters used for performance evaluation of the controller are integral of absolute error (IAE) and total harmonic distortion (THD) of the output grid voltage. Effectiveness of the controller has been compared with robust PID controller tuned by Cohen-Coon (C-C) method and negative imaginary (NI) theory based second order controller. It was found that the proposed control scheme shows 86.11% and 96.04% less IAE than C-C PID and NI based second order controller, respectively, for resistive load. Also, the THD of grid voltage was 0.30 % and 0.12% less than C-C PID and NI based second order controller respectively. Modeling and simulation of the system has been conducted in MATLAB/Simulink platform.