Application of machine learning controller in matrix converter based on model predictive control algorithm

Abstract

Finite control set model predictive control (FCS-MPC) algorithms are famous in power converter for its easy implementation of constraints with cost function than classical control algortihms. However computation complexity increases when swicthing state is high for converters such as matrix converter, multilevel converters and this impose a serious drawback to compute multi-step prediction horizon MPC algorithm which further increases the computation. To overcome the above said difficulty, machine learning based artificial neural network (ANN) controller for matrix converter is proposed. The training data for ANN controller is derived from MPC algorithm and trained offline with an accuracy of 70.3%. The proposed ANN controller shows a similar and better performance than MPC controller in terms of total harmonic distortion (THD), peak overshoot during dynamic change in reference current and dynamic change in load parameter and less computation with less execution time. Further, ANN controller for matrix converter is tested in OPAL-RT using hardware in-loop (HIL) simulation and showed that it outperforms MPC controller.

The implementation of an optimized neural network in a hybrid system for energy management

“…In hybrid energy systems, the variability of the power generated by the photovoltaic panels, the variability of the power demanded by the load system and the presence of critical loads requiring uninterrupted power supply are critical points to consider [11]- [14]. To guarantee a stable, reliable power supply, it's essential to integrate an intelligent energy management system.…”