A novel reflective intelligence optimizer with machine learning (RIO-ML) for parameter estimation of photovoltaic...

This paper presents a new Reflective Intelligence Optimizer with Machine Learning (RIO-ML) approach to estimate the parameters of solar photovoltaic (PV) equivalent circuit models, which are highly nonlinear and multimodal, and hence require efficient handling by conventional optimizers. RIO-ML combines three major components: a multi-leader social learning algorithm with personal-best reflective memory, machine learning-driven adaptive control of important parameters using Multi-Layer Perceptron models, and progressive Gaussian refinement with reflective boundary treatment for improved convergence and robustness. The performance of RIO-ML is tested on the standard RTC France solar cell model with three different model settings: Single Diode Model (SDM) with 5 parameters, Double Diode Model (DDM) with 7 parameters, and Triple Diode Model (TDM) with 9 parameters, for 30 independent runs for each scenario. RIO-ML obtains the minimum RMSE of 8.739710 × 10^- 4 A, 8.456760 × 10^- 4, and 7.7546980 × 10^- 4 A for SDM, DDM, and TDM models, respectively, with corresponding low mean RMSE values of 2.223109 × 10^- 3 A, 2.282687 × 10^- 3 A, and 1.717649 × 10^- 3 A. Comparative studies reveal that RIO-ML performs better than some of the best metaheuristic algorithms available in the literature with respect to solution quality, convergence rate, and robustness, while maintaining the maximum absolute current errors less than 1.6 × 10^- 3 A for all models. These findings clearly indicate that the developed RIO-ML approach is an effective and efficient tool for accurate estimation of PV parameter values.