Integration of predictive and computational intelligent techniques: A hybrid optimization mechanism for PMSM dynamics reinforcement

Abstract

This paper presents an integrated approach combining a sequential neural network (SNN) with model predictive control (MPC) to enhance the performance of a permanent magnet synchronous motor (PMSM). We address the challenges of traditional control methods that struggle with the dynamics and nonlinear nature of PMSMs, offering a solution that leverages the predictive capabilities of MPC and the adaptive learning potential of neural networks. Our SNN-MPC model is contrasted with state-of-the-art genetic algorithm (GA) and ant colony optimization (ACO) methods through a comprehensive simulation analysis. This analysis critically examines the dynamic responses, including current, torque, and speed profiles, of the PMSM under proposed hybrid control strategies. The heart of the work deals with the optimal switching states and subsequent voltage injection to the inverter fed PMSM drive by a predefined minimization principle of a current modulated objective function, where MPC constitutes an integral finite control set (IFCS) mechanism for voltage vector selection and thereby selects the optimized integral gains Kd and Kq for direct and quadrature axes, respectively, with the FCS gain Kfcs obtained from implemented intelligent techniques. Based on the control criteria, the SNN-MPC scheme was established as the preferred benchmark with optimized tuning values of Kd = 0.01, Kq = 0.006, and Kfcs = 0.13, as compared to the gain values tuned from GA and ACO. The experimental setup utilized MATLAB and a Python environment for robust and flexible simulation, ensuring an equitable basis for comparison across all models.

Keywords

• permanent magnet synchronous motor
• model predictive control
• genetic algorithm
• ant colony optimization
• sequential neural network
• voltage source inverter
• finite control set
• integral finite control set