Proactive TCN–Attention-LSTM Based MPPT Architecture for anticipating short-term irradiance and temperature variations for Photovoltaic System

Abstract

Maximum Power Point Tracking (MPPT) is an important process in photovoltaic (PV) systems to get the highest possible power from solar panels under changing weather conditions. Traditional MPPT methods such as Perturb & Observe (P&O) and Incremental Conductance (INC) are slow, cause small power losses due to oscillations, and do not perform well when sunlight changes quickly or when panels are partially shaded. Rapid variations in solar irradiance and photovoltaic (PV) module temperature significantly affect the operating point and power output of PV systems, thereby challenging the dynamic performance of conventional maximum power point tracking (MPPT) techniques. This paper proposes a Proactive Temporal Convolutional Network–Attention–Long Short-Term Memory (TCN–Attention–LSTM) based MPPT architecture for anticipating short-term variations in solar irradiance and temperature and enabling predictive maximum power point tracking. The proposed framework employs a TCN to extract multi-scale temporal features from historical environmental and electrical measurements, while an attention mechanism identifies the most relevant temporal information influencing future PV operating conditions. The LSTM network subsequently models long-term temporal dependencies to forecast short-horizon irradiance and temperature variations. These predicted parameters are integrated with the MPPT controller to estimate the forthcoming maximum power point and proactively regulate the DC–DC converter duty cycle. By initiating control action before substantial environmental changes affect the PV array, the proposed architecture aims to minimize tracking latency, reduce power fluctuations, and improve maximum power extraction.