Power prediction of a wind farm cluster based on spatiotemporal correlations

Accurate power prediction of wind farm clusters is important for safe and economic operation of power systems with high wind power penetration. Current superposition and statistical scaling methods used in wind power prediction systems do not fully consider the relationships among wind farms in a cluster, thereby leading to insufficient power prediction accuracies. To improve the power prediction accuracy of wind farm clusters, a new method based on spatiotemporal correlations is proposed herein. First, three correlation coefficients are used to represent spatiotemporal correlation characteristics of wind farms in a wind cluster. The Shapley value method is used to weight these coefficients, and a standard wind farm is found by combining the nominal capacities of the wind farms. Then, considering the spatiotemporal factors that affect wind power generation, a characteristic matrix of the wind farm cluster is constructed, and the key characteristics are extracted using a convolutional neural network (CNN). Considering the time series characteristics of wind power generation, a long and short term memory (LSTM) neural network is used to establish the mapping relationship between key characteristics and power generation, and power prediction of a wind farm cluster is performed. Finally, by utilizing the actual operating data of wind farm clusters in North China as an example, feasibility and effectiveness of the proposed method are verified. The proposed system provides a new high-precision method for future wind farm cluster power predictions.

Automated summary

This paper proposes a new method based on spatiotemporal correlations to improve the accuracy of power prediction in wind farm clusters by analyzing the relationships and characteristics among wind farms in the cluster and using neural networks for prediction. This method has been validated using actual operating data from wind farm clusters in North China, demonstrating its feasibility and effectiveness in providing a high-precision approach to future wind farm cluster power predictions.