Short-term Wind Power Forecasting Algorithm Based on Similar Time Period Clustering

Abstract

Wind power has strong randomness and volatility, In view of this, a short-term wind power forecasting algorithm based on similar time clustering is proposed. As the rotation of the blade has inertia, giving the definition of similar period by analyzing historical data and the optimal length of similar period by experiment. On this basis, we use K-means algorithm to cluster similar time. In the prediction process, the similarity period is divided by calculating the distance between it and the center of each cluster, then the optimal similar time series is selected. Taking the set as the training sample, the Elman neural network model is established by using the power curve and the meteorological information. This model is employed to iteratively predict wind power in future periods. The actual cases show that our method is an effective expansion of idea based on forecast of similar day, which can effectively mining the valuable information contained in historical data and improve the shortterm wind power forecast performance.