Very Short-Term Solar Power Forecast using Data from NWP Model

Abstract

This article presents a method to forecast solar power 4 hours in advance by using forecast weather data from Numerical Weather Prediction (NWP) and the measurement obtained from weather monitoring instrument with feature engineering by machine learning techniques using Feed-forward Neural Network (FNN) and Recurrent Neural Network (RNN). Four types of relevant variables—forecast weather data, the average of forecast weather data, forecast weather data with the measurement from weather instrument, and the average of forecast weather data from nearby areas with the measurement from weather instrument—are used to find the right model and input variable for the forecast. The result shows that by using Feed-forward Neural Network with the forecast weather data and the measurement from weather monitoring instrument, the Root Mean Square Error (RMSE) value is 8.13%. On the other hand, by using Feed-forward Neural Network with just the forecast weather data from nearby areas, RMSE is at 8.38% which is only 0.25% higher than the forecast with the measurement. Thus, only weather data from nearby areas can be used for prediction of the area without the measurement from weather monitoring instrument and we can use it for further prediction at the region level