Heuristic Binary Search for Modulated Predictive Control

Predictive Control optimization often relies on exhaustive enumeration of voltage vectors within the finite control set. Using Virtual Voltage Vector strategies tend to increase the number of predictions to enhance dynamic and steady state performance, leading to a higher computational burden. Modulated Model Predictive Control offers a solution by incorporating a modulation stage, but duty-cycle-based implementations can restrict the inclusion of constraints and nonlinearities in the cost function. This paper introduces the Heuristic Binary Search, an intelligent search algorithm that efficiently explores the solution space, accommodating arbitrary cost function constraints similar to Finite Control Set, while maintaining a fixed switching frequency. Applicable to various machine and inverter topologies, the proposed method is experimentally validated on a two-level voltage source inverter driving an induction motor using Predictive Torque Control and Deadbeat Robust Predictive Control. Experimental results comparing three variants of the Predictive Torque Control (one vector, three vector and modulated) show improvements in torque and flux ripple and improvements of current THD up to 30% over classic Modulated Predictive Torque Control implementation with reduced or similar computational cost. Results compared with the Finite Control Set Deadbeat Robust Predictive Control shows an improvement in THD of 40% with negligible impact in computational time.