Ultra-Short-Term Wind Power Forecasting Based on Fluctuation Pattern Clustering and Prediction

At present, wind energy has become one of the most promising renewable energy sources. However, large-scale wind power integration will adversely affect the power grid's safe and stable operation. So accurate wind power forecast is critically crucial for the power system to operate reliably and economically. Still, there are also many challenges obstructing it, such as the diversity of wind power fluctuation patterns. Focusing on this problem, a new forecast model based on fluctuation pattern clustering and prediction is proposed in this paper. First, wavelet decomposition is used to filter the original power time series. Then, according to swinging door algorithm and K-means clustering algorithm, the power fluctuation patterns are divided and clustered into three categories. The artificial neural network is used to predict the clustering labels of the fluctuation patterns. According to this result, the neural network is trained with different training sets classified by the tags to establish the power prediction model. The forecast accuracy of this model is better than that of direct prediction by the neural network. Finally, the actual data of five wind farms is used for experiments to verify the proposed model's feasibility and effectiveness.