Memory-based quadratic interpolation optimization with reinforcement learning for robust PV parameter estimation

Solar cell parameter extraction is a critical yet challenging multimodal optimization problem, directly impacting the efficiency and modeling accuracy of photovoltaic (PV) systems. The Quadratic Interpolation Optimization (QIO) algorithm, while possessing strong exploitation capabilities, is prone to premature convergence and lacks a robust exploration mechanism. To overcome these limitations, this paper proposes a Memory-based Reinforcement Learning QIO (MRQIO) algorithm. MRQIO integrates a reinforcement learning agent to dynamically balance exploration and exploitation by adaptively adjusting search weights based on population diversity and fitness improvement. Furthermore, a memory-based mechanism leverages historical high-quality solutions to guide the quadratic interpolation process, enhancing global search capability and preventing stagnation in local optima. The proposed algorithm was rigorously validated on 13 benchmark functions and five distinct PV models: RTC France single-diode (SDM) and double-diode (DDM), STM6-40/36, STP6-120/36, and PWP 201. MRQIO demonstrated superior performance, achieving the lowest Root Mean Square Error (RMSE) values of 0.000986 (SDM), 0.000987 (DDM), 6.7794E-05 (STM6-40/36), 0.000014 (STP6-120/36), and 0.00243 (PWP 201). Comprehensive statistical tests, including the Wilcoxon rank-sum and Friedman tests, confirmed that MRQIO’s performance is significantly more accurate and robust than state-of-the-art metaheuristics, establishing it as a powerful tool for high-precision PV parameter estimation.