基于机器学习的集群式风光一体短期功率预测技术

针对区域风、光电站群的功率预测，由于各站建站时间不同、单站预报精度残次不齐，导致传统的单站功率累加法预测精度和运行效率不高的问题，采用基于机器学习的二分K均值聚类算法分别对区域内的风电场和光伏电站群进行合理划分，结合区域内各电站历史功率数据及区域总历史功率数据的相关性，选取出各区域的代表电站。在对数值预报要素进行优化订正后，采用BP神经网络法建立基于风电场和光伏电站集群划分的短期功率预测框架模型。结果表明：采用该方法的集群式风电和光伏短期功率预测准确率高于或接近于传统单站累加的预测精度，且该方法在保证预测精度的同时，能够显著提高建模效率。

Short-Term Power Prediction for Wind Farm and Solar Plant Clusters Based on Machine Learning Method

Regional wind power and PV(photovoltaic) power forecast is an effective way to improve the robustness of power grid, however, the traditional single-station power accumulation method is poor in accuracy and operation efficiency, because the construction time of each station is different and the accuracy of single-station is various. Therefore, this paper proposes a method for short-term regional wind and PV power prediction based on feature clustering. Firstly, the machine learning-based Bisecting K-Means(BKM) clustering algorithm is used to reasonably divide the wind farms and PV stations in the region into clusters; Secondly, based on the correlation between of the historical power data of each power station and the total historical power data in the region, a representative power station is selected for each region; Thirdly, after optimizing and correcting the NWP(numerical weather prediction) model of each representative power station, a short-term power prediction framework model is established using BP neural network based on the cluster division of wind farms and PV power plants. The result shows that the proposed method is higher than or close to the traditional single-station power accumulation method in short-term prediction accuracy, but it can significantly improve the modeling efficiency while ensuring the prediction accuracy.