Design and implementation of predictive current controller based on artificial neural networks for voltage source inverters

In this study, five different Artificial Neural Network (ANN) based predictive current controller designs were made for voltage source inverters. In the first design, the designed static ANN-based current controller is trained by using the data obtained from the finite control set model predictive control (FCS-MPC) method using the mathematical model of the system Then, a feedforward reference current predictor ANN (RefPNN) was designed to predict sinusoidal reference current in the other four methods. Using this structure, two inverter simulations containing a hysteresis current controller were created, and two hysteresis-based predictive ANNs (Hist-PNN1 and Hist-PNN2) current controllers were trained with the data obtained from this simulation. Similarly, data from a simulation containing RefPNN and proportional-resonant (PR) current controller were used in training the fourth PR-based predictive current controller (PR-PNN). The fifth ANN-based predictive current controller (PI-PNN) training data are taken from a simulation which RefPNN and, as the current controller, a proportional-integral (PI) controller. A three-phase 5kVA inverter circuit with a 7MBP50RJ120 IPM module in the power stage and STM32f407 as a controller was designed for the experimental study. Five different ANN current controllers trained using offline data were tested both in the simulation environment and experimentally under different test conditions, and their results were compared. It has been determined that all methods have predictive properties and can be applied successfully.