This study introduces an adaptive artificial neural network (ANN) -based control system to enhance the efficiency of stand-alone photovoltaic (PV) systems under dynamic environmental conditions. Traditional maximum power point tracking (MPPT) methods, such as perturb and observe (P&O) and incremental conductance (INC), are hindered by slow convergence and oscillations. The proposed approach utilizes a hybrid ANN architecture with hyperbolic tangent (tanh) and rectified linear unit (ReLU) activation functions in a 6-3 neuron hidden layer structure, enabling real-time prediction of the optimal voltage (Vₘpp). Integrated with a PID-controlled DC-DC boost converter, the system seamlessly transitions between the solar harvesting, battery charging, and load supply modes. Trained on 10, 000 environmental samples (irradiance: 150–1000 W/m² and temperature: 25–50°C) using the Levenberg-Marquardt algorithm, the ANN achieved 99. 2% tracking accuracy with a mean squared error (MSE) of 1. 73×10⁻⁵ in 200 epochs. MATLAB/Simulink simulations demonstrated superior performance, surpassing P&O by 4. 1% and INC by 3. 2%, while maintaining a voltage ripple below 1. 5%. Key innovations include the hybrid ANN design that mitigates saturation effects, adaptive PID tuning for minimal oscillations, and a three-mode converter that ensures a stable 24 V load voltage during irradiance fluctuations. This work underscores the potential of machine learning in advancing renewable energy systems, offering a computationally efficient and hardware-ready solution for off-grid applications with enhanced reliability and precision.
Al-Husban et al. (2025) studied this question.