Ultra-Short-Term Wind Power Forecasting Model Based on Time-Section Fusion and Pattern Classification

Accurate wind power prediction can not only effectively improve wind power consumption capacity, but also improve the stability and economy of power grid operation. However, due to the fluctuation and randomness of wind speed, it is very difficult to predict accurately wind power at present. In addition, each single algorithm model has its own inherent limitations. Therefore, none of the above algorithms can always maintain the best prediction effect in different months. For the above reasons, it is very difficult for the wind farm to choose the optimal prediction model under different prediction scenarios. Finally, for the problem that a single algorithm can’t adapt to the forecasting scenarios of different months in the whole year, a time-section fusion pattern classification based ultra-short-term wind power forecasting model is proposed, which includes multiple forecasting models based on different machine learning theories, fusion mode classification model, and fusion model corresponding to each fusion mode. Firstly, different machine learning models such as artificial neural network (ANN), support vector machine (SVM), long and short-term memory neural network (LSTM) and autoregressive integrated moving average (ARIMA) are used to predict wind power with a single model. Secondly, the model of fusion pattern classification and recognition is trained and applied to fusion pattern recognition in different time periods. Finally, the final forecasting results are obtained by using the fusion model corresponding to the data fusion model. The simulation results of real data show that compared with the standard ANN model, SVM model, LSTM model and ARIMA model, the time-section fusion based ultra-short-term wind power ensemble forecasting model proposed in this paper has higher accuracy.