A new multistage short-term wind power forecast model using decomposition and artificial intelligence methods

Abstract

In this study, a new forecast model consist of three stages is proposed for the next hour wind power. In the first stage, wind speed, wind direction, and wind power have been forecasted by using historical data. Artificial Neuro-Fuzzy Inference System (ANFIS), Artificial Neural Network (ANN) and Support Vector Regression (SVR) have been chosen as forecast methods, while Empirical Mode Decomposition (EMD) and Stationary Wavelet Decomposition (SWD) methods have been preferred as pre-processing methods. The other two stages have been used to improve the wind power forecast value obtained at the end of the first stage. In the second stage, the forecast values found in the first stage have been applied to the same forecast methods, and wind power forecast value has been updated. In the third stage, a correction process is applied, and the final forecast value is obtained. While four-year data are selected as train data, two-year data are tested. SWD-ANFIS has given the best results in the first stage while ANN has given the best result in the second stage. Finally, the ensemble result has been found by taking the weighted average of the results of the three methods. Mean Absolute Error (MAE) values found at each stage are the 0.333, 0.294 and 0.278, respectively. The obtained results have been compared with literature studies. The results show that the proposed multistage forecast model is capable of wind power forecasting efficiently and produce very close values to the actual data.