Enhancing solar photovoltaic cell parameter estimation by a linear regression-based optimization technique

Abstract Accurate mathematical modeling and optimal parameter extraction are important for improving the efficiency of solar photovoltaic systems, financial planning, and manufacturing. A novel linear regression-dynamic inertia particle swarm optimization (DIPSO) algorithm has been developed in this study. Linear regression forms the backbone of this model by offering strategic initialization of key parameters, i.e. photogenerated current and series resistance. The initial values from linear regression are optimized by accelerating the convergence rate quickly and decreasing the computational time during the optimization process, whereby the optimization algorithm attains spectacular speeds of 0.397 s for the KC 200GT photovoltaic module and 0.271 s for the RTC-France solar cell for the triple-diode model. The linear regression helps in the initialization of the parameters near the optimal values, thereby minimizing the search space for the subsequent particle swarm optimization with two significant improvements: Dynamic inertia weight adjustment and velocity clamping. Dynamic inertia adjustment accelerates convergence by focusing the search on promising regions, whereas velocity clamping stabilizes the movement of particles to achieve efficient exploration with minimal computational time. The proposed linear regression-DIPSO approach is applied to the KC200GT photovoltaic module at standard test conditions and to the RTC-France solar cell at 33°C, 1000W/m². The lowest root mean square error achieved is 2.357 × 10-4 and 8.65 × 10-4 for KC200 GT and RTC-France for the triple-diode model, respectively. A comparative study establishes that the proposed linear regression-DIPSO approach surpasses conventional algorithms like simulated annealing and DIPSO in achieving faster convergence without compromising the quality of solutions. This research not only opens new avenues for effective parameter extraction in different photovoltaic models but also reveals the key benefits of integrating machine learning and optimization techniques for next-generation solar energy applications.