A novel decoupled identification-and-tracking framework for photovoltaic global maximum power point tracking under complex shading scenarios

Partial shading in photovoltaic (PV) arrays introduces nonlinearities that create multiple local maxima in the power–voltage curves, often causing conventional and metaheuristic trackers to converge on suboptimal peaks. To address this, this paper proposes a novel physics-informed hierarchical framework that decouples the tracking process into two distinct stages: global identification and localized tracking. In the identification phase, a distributed Physics-Informed Neural Network (PINN) architecture estimates the maximum power point voltage of individual modules using only local electrical measurements, eliminating the need for costly irradiance or temperature sensors. A hierarchical fusion rule is then introduced, leveraging the physical insight that the array-level Global Maximum Power Point (GMPP) is primarily governed by the highest-power string. The hierarchical fusion rule enables precise coarse localization of the GMPP region. In the tracking phase, this localization constrains a Guided Salp Swarm Optimization (GSSA) algorithm to a narrow voltage window, significantly accelerating convergence and preventing local-peak entrapment. The generated reference voltage is enforced by a Robust Adaptive Integral Backstepping Sliding-Mode Controller (RAIBSMC), ensuring stable and rapid regulation despite parametric uncertainties. Simulation results under complex partial shading scenarios show that the proposed framework identifies the global maximum power point region with up to 99.27% voltage estimation accuracy and achieves convergence within approximately 5 ms. Experimental validation on a microcontroller-based platform further confirms the real-time feasibility of the method, with the proposed framework tracking dynamic shading transitions with an experimental convergence time of approximately 6 ms.