Genetic Algorithm for PI Controller Design of Grid-connected Inverter based on Multilayer Perceptron Model

Abstract

Proportional-integral (PI) controllers are the most commonly used units in the control system of power electronics converters. The design of PI controller has long been an empirical process and is usually a time-consuming task. In this paper, the raw dataset is generated from a grid-connected inverter case running on Real Time Digital Simulators (RTDS). A clean training dataset is used to train a multi-layer perceptrons (MLP) neural network, which can approximate the mapping between PI parameters and the performance of the grid-connected inverter. Based on the verified MLP model, genetic algorithm (GA) is applied to design PI parameters for improving the active power control of the grid-connected inverter. The PI parameter tuning procedure requires almost no human interference. The validity of the proposed method in designing PI parameters is verified by simulation results. The benchmark case uses PI parameters from RTDS’s tutorial without optimization. The active power control performance of the grid-connected inverter is improved by using the optimized PI parameters.