A Modified Neural Network for Predicting the Solar Photovoltaic Power Generation Using a Weather and Operational Parameter

Abstract

Solar PV systems are facing issues with feeding power to the local grid due to the unreliability of power generation. Though the solar PV system power generation is not stable, it is predictable to maintain the grid stability. In this study, a modified neural network is developed to predict the power generation for a 500-kW solar farm under Thailand's climatic conditions. Year-round solar PV plant operational data are used to train the forecasting model and over 15% of the period is used to predict the power generation and validated with the actual power profile. It is found that 0.22 kW of average power generation difference is noted from the forecasting model as compared to the actual power profile which concludes that the forecasting is accurate with 99.86% over the testing period. Overall, a 2.88 kW to -4.67 kW difference is noted between the actual and precited power and the corresponding MAE, MSE, and RMSE attained 0.87, 1.32, and 1.15, respectively. Further, it is concluded that the developed ANN forecasting model is highly recommended for commercial purposes to avoid penalties from grid authority as well as to improve grid stability.