Solar PV Generation Forecast Model Based on the Most Effective Weather Parameters

Abstract

Solar energy is one of the major renewable energy sources with the potential to cope with the future energy challenges. But the penetration of solar PV generation in the electrical grid is a serious concern because of variable availability. Therefore, solar PV generation forecasting is essential for planning and efficient operation. The forecasting model is based on Artificial Neural Network (ANN) with forecasted and historical weather parameters i.e., temperature, dew point, relative humidity and wind speed as inputs. The aim of this study is to determine the most effective combination of weather variables to be used as input to the model. For this, all the possible combinations of the inputs are applied to ANN and the best one is obtained by analysis of the results. Mean Absolute Percentage Error (MAPE) is used as a measure to compare the results. To train the ANN model, one year's weather and generation data of 20.8 kW PV system with an hourly resolution is used. 24 hours ahead forecasting of the generation is done using forecasted weather data of 14 days selected from the dataset of 130 days. Combination of three parameters (temperature, relative humidity and dew point) results in an average MAPE of 14.86% while the use of all four parameters as inputs gives 14.33% of MAPE.