Abstract
Abstract The occurrence of partial shading (PS) in photovoltaic (PV) systems has a profound impact on their operational performance, as it triggers the formation of hotspots, ultimately resulting in localized heat and potential irreversible damage to the PV modules. In response to these critical issues, this research presents a three-step framework based on voltage time-series analysis for the detection, classification, and prediction of PS in PV modules. The framework combines rule-based detection algorithms with advanced deep temporal learning architectures, utilizing cost-effective Internet of things (IoT) sensors to mitigate reliance on external imaging systems or complicated weather setups. Various shading scenarios were emulated to investigate the impact of PS on the PV module, ranging from single-cell to full-module shading under the same environmental conditions. Experimental validation reveals outstanding performance across all operational stages: 99.05% detection accuracy; 95.51% classification precision for dynamic and static shading patterns (SPs); and superior prediction capability with an R² value of 0.932 and a Mean Absolute Error (MAE) of 1.349 V for short-term intensity forecasting. The proposed approach offers two advantages of real-time monitoring and precise shading classification, thereby enabling a transition from conventional reactive partial shading detection (PSD) strategies toward proactive and predictive management while enhancing energy efficiency and operational reliability under dynamic environmental conditions.