Short-term Power Prediction of Wind Power Cluster Based on KOA-PLE

With the gradual acceleration of the world's energy transformation, the construction of new energy is becoming more and more large-scale. At the same time, the proportion of new energy in the power grid is increasing, and the high volatility of wind power poses certain challenges to the stability and reliability of the power system. Therefore, accurate large-scale wind power prediction is very necessary. This paper presents a multi-task learning predicting method for wind power clusters. Firstly, the wind power cluster is divided into subgroups with similar output characteristics, and then the different subgroup power is used to predict the multi-task targets. Meanwhile, the Kepler optimization algorithm (KOA) is used to optimize the model parameters. When the method is applied to a wind power cluster in China, RMSE and MAE are reduced by 2.07% and 1.92% compared with the traditional multi-task learning method, which proves the effectiveness of the proposed method.