A Short-Term Power Prediction Method for Wind Farm Cluster Based on Time Dynamic Graph Network and Improved Transformer Model

In recent years, wind power has gradually replaced thermal power as the main source of new energy generation. Wind power prediction in advance can help formulate scheduling strategies and improve the stability of the power system. With the advent of the Industry 4.0 and the widespread application of artificial intelligence technology, artificial algorithms have been extensively used in the field of wind power prediction. To further explore the spatiotemporal correlation of every wind farm in the wind farm cluster, a temporal dynamic graph network is used to divide the wind farm farm into several sub-clusters. Then, the KAN-BiLSTM-Transformer algorithm is used to predict wind power of each wind sub-cluster. Finally, the wind power of each sub-cluster is added to attain the final predicted power of the wind farm cluster. After verification, the RMSE and MAE of the proposed method is 0.06849 and 0.05805 respectively, with an accuracy of 93.15%, which is superior to the compared algorithms. Proposed method can provide more accurate prediction results.