Intelligent LSTM-Based Prediction and Adaptive Barrier Sliding Mode Control With Metaheuristic Optimization for Maximum Power Point Tracking in Photovoltaic Systems

The rapidly growing shift towards the sustainable energy source has placed the photovoltaic (PV) systems as a keystone of the present-day renewable energy production. However, a high nonlinearity of the system and the very rapid change of the environmental conditions mean that preserving their operation at the actual maximum level of power is still a great challenge. In a bid to address such shortcomings, this paper proposes an intelligent maximum power point tracking (MPPT) architecture that combines a predictive Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) and an Adaptive Barrier Condition-based Super-Twisting Sliding Mode Controller(ABC-STSMC). To achieve the maximum possible PV reference voltage, the LSTM-RNN component tries to predict the temporal dependencies of irradiance and temperature trends, and to guarantee the finite time convergence, high disturbance suppression, and minimized chattering features, the ABC-STSMC incorporates adaptive gain control and a barrier-type stabilization. The controller parameters are autonomously optimized using the Whale Optimization Algorithm (WOA), enabling self-tuning capability and consistent dynamic performance across varying operating conditions. For comparative evaluation, a Condition-Based STSMC (C-STSMC) is also implemented. Results from MATLAB/Simulink and Controller Hardware-in-the-Loop (C-HIL) validation reveal that the WOA-tuned ABC-STSMC achieves a rise time of 12 ms, settling time of 19 ms, overshoot below 1.5%, and steady-state voltage error within 0.2 V, sustaining MPPT efficiency above 97%. Relative to C-STSMC, Fuzzy Logic Control, and the Perturb and Observe (P&O) algorithm, the proposed controller enhances convergence rate by up to 36% and suppresses voltage oscillations by over 40%, even under partial shading and dynamic load variations. These findings affirm that the synergistic combination of predictive neural modeling, adaptive nonlinear control, and evolutionary parameter optimization delivers a robust and scalable solution for highefficiency PV energy extraction in real-world uncertain environments.