Short-Term Wind Power Forecasting Based on Multi-Scale Fusing Deep Learning

Abstract

An accurate forecast of wind power plays an increasingly important role in economic dispatch and power system operations. Nevertheless, the randomness, intermittence, and volatility of wind power pose great challenges in power system balance. To improve the forecasting accuracy of wind power, a multi-scale fusing deep learning method is proposed to forecast short-term wind power for multiple wind turbine generators in a wind farm. The inherent spatio-temporal characteristics are studied by dividing the time scale into a small-scale nearby part and a large-scale historical part. The fine-grained dynamic characteristics in the small-scale nearby part are extracted using stacked three-dimensional convolutional neural networks (3D-CNN), and the steady-state characteristics in the large-scale historical part are extracted using Spatio-Temporal Long Short-Term Memory (ST-LSTM) units. Then, wind power for multiple wind turbine generators can be forecasted by fusing the two deep learning channels. An actually collected wind power dataset is used to validate the effectiveness of the proposed method, which is exported from 33 wind turbine generators in a wind farm located in a northwestern province of China. The experimental results show that the proposed wind power forecasting method can improve the forecasting accuracy effectively when compared with the state-of-the-art wind power forecasting methods.