Federated leveraging AI-assisted MPPT for real-time photovoltaic performance optimization using machine learning

Maximum power point tracking is a necessary power optimization technique for maximizing the energy output of photovoltaic PV systems that operate in variable environmental conditions. While classical MPPT algorithms, such as Perturb and Observe and Incremental Conductance (IncCond), are well implemented, research is now underway on artificial intelligence techniques to improve tracking performance. This paper reports on a control-oriented benchmarking study of classical and AI-assisted MPPT strategies within a unified discrete-time simulation framework implemented in MATLAB. An artificial neural network is proposed in an AI-assisted MPPT architecture to estimate the optimal PV operating voltage, while a conventional proportional-integral voltage controller enforces it and maintains closed-loop stability. The performance of P&O, IncCond, and the AI-assisted MPPT is studied under uniform irradiance steps, fast irradiance fluctuations, and partial shading conditions. Under a 1000->600 W/m² irradiance step, IncCond delivers a tracking efficiency of 99.89% with very low power ripple (0.0014 W), which is significantly better than P&O (98.13%) and the AI-assisted approach (67.92%). In partial shading, P&O and IncCond have efficiencies of 96.21% and 92.99%, respectively, while the AI-assisted MPPT has an efficiency of 67.84%. These results show that the IncCond is a strong and reliable baseline, and that the AI-assisted MPPT offers valuable insight into hybrid control design and requires careful, control-aware integration.