Hybrid and Ensemble Methods of Two Days Ahead Forecasts of Electric Energy Production in a Small Wind Turbine

摘要: The ability to forecast electricity generation for a small wind turbine is important both on larger scale where there are many such turbines (because it creates problems networks managed by distribution system operators) and prosumers allow current energy consumption planning. It also owners of systems in order optimize the use various sources facilitate storage. research presented here addresses an original, rarely predicted 48 h forecasting horizon turbines. This topic has been rather underrepresented research, especially comparison with forecasts large farms. Wind speed used as input data. We have analyzed available data identify potentially useful explanatory variables models. Eight sets increasing amounts were created analyze influence types quality. Hybrid, ensemble single methods predictions, including machine learning (ML) solutions like long short-term memory (LSTM), multi-layer perceptron (MLP), support vector regression (SVR) K-nearest neighbours (KNNR). Original hybrid methods, developed specific implementations based methods’ decreased errors methods. “artificial neural network (ANN) type MLP integrator methods” method incorporates original combination predictors. Predictions this lowest mean absolute error (MAE). In addition, paper presents method, called “averaging without extreme forecasts”. root square (RMSE) among all tested LSTM, deep network, best second one, while SVR, KNNR and, especially, linear (LR) perform less well. prove that lagged values forecasted time series slightly increase accuracy predictions. same applies seasonal daily variability markers. Our studies demonstrated using full set proposed yield error. applicable other short-time renewable (RES), e.g., photovoltaic (PV) or hydropower.