ARTIFICIAL INTELLIGENCE-OPTIMIZED ADAPTIVE PULSE WIDTH MODULATION CONTROL FOR GRID-CONNECTED PHOTOVOLTAIC INVERTERS

Unofficial AI-generated summary based on the public title and abstract. Not an official translation.

🔬研究者:The reinforcement learning-based adaptive PWM method offers a novel approach for real-time optimization of inverter control, contributing to power electronics and renewable integration research.

🏢実務担当者:PV inverter manufacturers can implement this AI-optimized PWM to achieve higher efficiency and better power quality, with a quantified payback period for investment decisions.

ABSTRACT : The increasing penetration of grid-connected photovoltaic (PV) systems, particularly in three-phase applications, has intensified the need for inverter control strategies that simultaneously maximize energy efficiency and ensure high power quality under variable operating conditions. The problem. Conventional Pulse Width Modulation (PWM) control strategies for three-phase grid-connected PV inverters face an inherent trade-off between energy efficiency and power quality. Fixed and adaptive PWM techniques often struggle to maintain optimal performance across changing irradiance levels and grid conditions, while advanced methods frequently increase system complexity without delivering consistent gains in both efficiency and Total Harmonic Distortion (THD). Goal. The goal of this study is to conduct a comprehensive comparative analysis of widely used PWM control strategies and to develop an Artificial Intelligence–Optimized Adaptive PWM (AIO-PWM) algorithm capable of dynamically balancing efficiency and power quality in three-phase PV inverter systems. Methodology. The proposed AIO-PWM algorithm is benchmarked against four conventional methods: Multi-Objective Adaptive PWM (MOA-PWM), Fixed Frequency PWM, Hysteresis Current Control, and Predictive Current Control. AIO-PWM employs reinforcement learning–inspired techniques to dynamically optimize the switching frequency and modulation index based on real-time operating conditions. The control framework evaluates multiple candidate solutions using a weighted multi-criterion scoring system that balances efficiency and THD objectives. Online learning mechanisms track historical performance data, while exploration–exploitation strategies enable continuous adaptation to changing conditions. Performance evaluation is carried out using real solar irradiance data obtained from NASA POWER for a North African location (32.49°N, 3.67°E) in a three-phase simulation environment. Results. Simulation results show that AIO-PWM achieves an average efficiency of 92.68%, outperforming the MOA-PWM baseline (91.76%) by 0.92%, and delivering an additional 11.35 kWh of energy compared to Fixed PWM. The proposed method maintains a competitive THD of 3.02% and achieves the highest Power Quality Factor (57.77) among all evaluated algorithms. Scientific novelty. The scientific novelty of this work lies in the integration of reinforcement learning–inspired adaptive optimization within a PWM control framework, enabling continuous online tuning of control parameters through historical performance evaluation and multi-objective decision-making in three-phase PV inverter applications. Practical value. Economic analysis indicates a 1.8-year payback period for AIO-PWM compared to 0.6 years for MOA-PWM, justifying the additional investment for premium three-phase applications where both efficiency and power quality are critical. The results provide practical guidance for selecting PWM control strategies based on technical performance and economic considerations in three-phase grid-connected PV systems.

• Zenodo https://zenodo.org/records/20832484first seen 2026-06-25 04:22:15 · last seen 2026-06-26 04:28:58