Lightweight Machine Learning-based Auto-Tuning of FCS-MPC for CSC Multilevel Inverters

Model Predictive Control (MPC) has become a widely adopted control technique for Multilevel Inverters due to its ability to manage multi-objective optimization problems under system constraints. However, a key challenge in MPC implementation lies in selecting appropriate weighting factors for the cost function, as fixed values often lead to suboptimal performance under dynamic operating conditions. Thus, this paper presents a lightweight auto-tuning method for the voltage weighting factor in Finite Control Set MPC (FCS-MPC), applied to a single-phase grid-connected 9-level Crossover Switches Cell inverter. The proposed approach employs low computational complexity machine learning models, Linear Regression and Support Vector Machine, trained offline on a minimal dataset comprising the DC link voltage and reference current. These models are embedded into the control loop to enable real-time adjustment of the voltage weighting factor. The presented comparative simulation results confirm the effectiveness of the proposed technique across a wide range of operating conditions. Compared to more complex AI-based solutions, this work contributes a simple yet effective ML-based tuning strategy that improves control performance with minimal computational overhead.