Deep Learning-Based MPPT for PV Systems: LSTM Forecasting and Adaptive TSMC via PPO Agent

Abstract

Maximum power point tracking (MPPT) is a crucial technique for maximising the efficiency of photovoltaic (PV) systems by continuously adjusting the operating point of the PV array to extract the maximum possible power under varying environmental conditions. Conventional MPPT methods are often simple to implement but suffer from slow response times, oscillations, and reduced efficiency under rapidly changing irradiance and temperature conditions. Advanced control strategies like sliding mode control (SMC) offer improved robustness but can still exhibit chattering effects and may not be optimal for the dynamic, non-linear behaviour of PV systems. The growing complexity of PV systems, particularly those operating under partial shading, necessitates a more intelligent and adaptive approach that can predict optimal operating points and adjust control parameters in real time to overcome these limitations. This paper proposes an innovative hybrid approach for MPPT in PV systems, combining a long short-term memory (LSTM) neural network for dynamic voltage prediction, a Terminal Sliding Mode Control (TSMC) control for robust tracking, and a proximal policy optimisation (PPO) reinforcement learning (RL) agent to optimise TSMC parameters in real-time. This synergy combines LSTM’s predictive capability with deep reinforcement learning (DRL)’s adaptive optimisation creating a more efficient and robust MPPT system than conventional methods. Simulations under various environmental conditions show that the proposed hybrid controller significantly outperforms a conventional sliding mode controller and a standalone TSMC system. Notably, the integrated deep learning (DL) and RL approach enhances overall energy efficiency. The proposed method achieves a superior efficiency of 97.4%, a lower average error of 0.25, and a faster response time of 65 ms compared to the conventional methods. This integration of DL allows for dynamic adaptation to irradiance and temperature variations, improving the overall energy efficiency of the PV system.