Machine Learning Based Adaptive Prediction Horizon in Finite Control Set Model Predictive Control

Abstract

In this thesis, an adaptive prediction horizon approach based on machine learning is presented for the nite control set model predictive control (FCS- MPC) of power converters. Normally, in FCS-MPC, the prediction horizon (N) is xed during both the transients and the steady-state. A large N improves performance while signi cantly increasing the computational cost. A novel technique is presented where the prediction horizon (N) adapts to the instantaneous states i.e. il, vo and vo;err = (vref ð€€€ vo). Simulations are run for di erent combinations of N, il, vo and vref to create a data set of the optimal N against the instantaneous states and the error between the reference output voltage and the output voltage and an arti cial neural network is trained to tell the optimal length of N. The proposed method is simulated on a boost converter in Simulink. The simulation results show that the computational cost is reduced without a ecting the performance.