Short-Term Prediction of Wind Farm Cluster Power Based on Load Risk Scenarios and Improved Weighted Loss During Peak and Valley Load Periods

Load peak and valley periods as the key scheduling scenarios, the effectiveness of wind power prediction becomes extremely important, in order to minimize the impact of wind power on the power system due to prediction errors, a short-term prediction method of wind farm cluster power considering risk scenarios and improved loss function is proposed. Firstly, the coupled correlation mechanism between meteorological type and load demand is explored, and the load risk scenario and meteorological correlation are derived. Second, according to the risk scenarios of load demand, the traditional loss function is improved and used to train the power prediction model under the risk scenarios. Then, a dynamic division strategy of wind farm clusters considering the meteorological similarity of the previous day is proposed to obtain the sub-clusters of similar output. Finally, the model with improved weighted loss function is used to predict the wind power for different regional sub-clusters and extended to wind farm clusters. The above method is validated using the wind farm clusters in West Inner Mongolia, China. And the results show that in the low-risk load scenario, the accuracy of the model with improved loss is higher, and the accuracy decreases with the increase of the load risk level, but the overall risk indexes are significantly lower than the method with traditional loss function, which proves the validity of the proposed method.