Design of Neural Network-based MPPT with Neutral Point Clamped Converter for Grid Connected PV System

Abstract

Energy is now becoming one of the key factors of modern civilization. In our daily lives, it enables comfort and sustainability, despite the electricity demand still facing challenges in generation. Machine learning (ML) techniques are becoming crucial for optimizing the performance of solar energy harvesting systems to overcome this problem. The main purpose is to develop a model for a supervised ML-based maximum power point tracking (MPPT) control system utilizing a neutral point clamped (NPC) converter, ensuring a reliable power supply under varying operational and environmental conditions. This study investigates a neural network (NN) based Levenberg-Marquardt (LM) train algorithm to extract the MPPT power generation in the grid-connected solar photovoltaic (PV) system with the NPC converter model and explore the performance of energy management systems. Utilizing MPPT, we enhance the efficiency of the proposed solar system. It investigates the battery operation strategies under different loading conditions. To evaluate the NPC converter, we suggested a space vector pulse width modulation (SV-PWM) technique of 15% to enable control of the DC link neutral point voltage, thereby adjusting switching states and reducing operational loss. The system performance demonstrates high tracking efficiency, reduces oscillation, facilitates smooth load switching, and improves power quality with an effective energy storage system. The results indicate that the Hybrid (P&O, NN) MPPT is superior (96% efficiency) at energy harvesting, while the outperformed LM-based NN MPPT is better (SOC range ~ 45.0042%) at energy storage. The proposed framework offers a reliable and efficient solution for modern grid-connected PV systems with energy storage.