Simplification of Weather Parameters for Solar PV Power Forecast Model Using Machine Learning Algorithms /

Abstract

With the increase of energy demand, solar energy is the most reliable and eco-friendly alternative to the depleting conventional energy resources. But the intermittent nature of solar energy resources poses a great challenge for the research community to rely on this technology completely. This is because the variable and time varying weather parameters affect the output power generated by solar power plant. This uncertainty of the generated power is very hazardous and dangerous for the power system as it causes severe stability issues in the system. To meet this uncertain behavior of the solar power system, an accurate forecasting model is a necessity to predict the output power of PV power plant. This research is serving the purpose of forecasting the day-ahead power output of the PV power plant located at Lahore, Pakistan. Three machine learning techniques i.e. Extreme Gradient Boosting (XGB), Support Vector Machines (SVM) and Artificial Neural Networks (ANN) are used for this purpose by applying the input weather parameters obtained from an online site. These models are trained by using the recorded 1-year data of weather parameters and the power generated data by the power plant with an hourly resolution. The test set of the model is obtained by selecting the particular samples from the forecasted weather data of 180 days. These samples are selected on the basis of seasonal and weather variations. The model is applied on the test set to predict the power output for the selected samples. The models are evaluated using the evaluation metrices i.e. Root Mean Squared Error (RMSE) and Normalized Root Mean Squared Error (nRMSE) which compare the predicted power and the actual power obtained for the input selected samples. These results are then compared to analyze the trend of errors in sunny and rainy days of winters and summers for all the techniques used. After the analysis, it is observed that XGB outperform the rest of the techniques in terms of accuracy, efficiency, time conservation and complexity. For XGB, SVM and ANN, the overall nRMSE obtained for the selected samples is 12.5%, 14% and 15% respectively.