Machine Learning for Model Predictive Control of Cascaded H-Bridge Inverters

Abstract

Finite Control Set Model Predictive Control (MPC) is an effective control technique for Cascaded H-Bridge converters. However, its computational complexity becomes impractical when the number of levels of the converter increases. Machine Learning techniques can be successfully used to reduce the computational burden of the optimal control computation and this paper provides a comparison among conventional MPC and well-known Machine Learning techniques: Support Vector Machines, Regression Trees, Neural Networks and Linear Regression. A simulation study is presented for a Cascaded H-Bridge Static Synchronous Compensator varying the number of levels and using different Machine Learning control strategies. The results underline that some ML techniques can substantially reduce computational complexity while keeping the performance comparable with the optimal control.