Performance Evaluation of Si, SiC, and GaN Based Boost Converters for Photovoltaic Applications Using MPC-CNN Duty Cycle Optimization

This paper presents a novel hybrid control strategy for DC-DC boost converters in Photovoltaic (PV) systems, combining Model Predictive Control (MPC) with Convolutional Neural Networks (CNN) to optimise the duty cycle in real time. The MPC–CNN framework leverages MPC’s predictive accuracy and constraint-handling with CNN’s fast inference and adaptability, ensuring robust voltage regulation under varying irradiance and load conditions. The hybrid scheme dynamically selects between CNN and MPC outputs based on real-time performance metrics, balancing response speed and control precision. Simulations with a 100-kW dynamic Photovoltaic (PV) profile demonstrate that MPC achieves near-zero steadystate error, while CNN provides faster transient responses. The hybrid controller surpasses both in maintaining voltage stability and energy efficiency. Additionally, a performance comparison of semiconductor technologies—Silicon (Si), Silicon Carbide (SiC), and Gallium Nitride (GaN)—shows that GaN-based converters achieve the best results, with 97.6% efficiency, 2.8 V ripple, 6.4 W switching loss, and the fastest transient response of 0.17 ms. These findings confirm the effectiveness of the CNN-enhanced MPC approach and establish GaN’s superiority for compact, high-performance PV applications. Future work will explore real-time embedded implementation, adaptive CNN retraining, and integration into smart grid and energy management systems.