Solar photovoltaic power prediction using artificial neural network and multiple regression considering ambient and operating conditions

This paper proposes artificial neural network (ANN) and regression models for photovoltaic modules power output predictions and investigates the effects of climatic conditions and operating temperature on the estimated output. The models use six days of experimental data creating a large dataset of 172,800 × 7. After data preprocessing, the appropriate attributes were selected as inputs and taken into account as features; solar irradiation, ambient air and module temperature, wind speed, and relative humidity, while the power generation as a target. In light of these data, the effect of training algorithm on the predictive performance of the ANN model was investigated. Results show that solar irradiation, ambient and module temperatures are key factors in predicting PV module power generation, as these variables are strongly correlated with PV power output. Moreover, the Levenberg-Marquardt algorithm was found to be the best training procedure. The ANN model demonstrated higher accuracy than the developed multiple linear regression models. However, the proposed Rational-Power-Law (RPL) and Power-Law (PL) models were able to capture the nonlinearity in the system, as assessed by coefficient of determination (R2) and the Mean Absolute Error (MAE), and successfully supplied a very high level of precision. The ANN, and both RPL and PL models provided comparable performance, attaining an R2 of 0.997, 0.998 and 0.996, and a MAE of 1.998, 1.156, and 1.242, respectively, when compared to experimental results. Furthermore, models proposed in this study were evaluated and compared with others available in literature and have demonstrated superior performance and better accuracy.