基于BR-SOM聚类算法的配电网短期负荷预测

Research on Short-term Load Behavior of Distribution Network Based on BR-SOM Clustering Algorithm

为了提高配电网短期负荷预测能力,建立贝叶斯正则化(Bayesian regularization(BR))的自组织特征映射网络(SOM)聚类算法对配电网短期负荷用电行为分析模型。输出层拓扑内存在紧密关联的节点与邻域节点,能够根据各自的特点开展学习,因此,相邻节点将会形成相近的权重,并达到与相近输入节点的良好匹配性。之后选择宁夏某地区的电网作为测试对象对该方法进行有效性验证。研究结果表明:通过改进后的Relief算法对电力系统负荷特征向量实施筛选,最终确定5个,大幅降低了特征量的数量,达到简化负荷描述的效果。测试了存在显著特征的三类用户负荷平均准确率,SOM算法准确率均在97%以上,验证了这个模型的准确性。

In order to improve the short-term load prediction ability of the distribution network,a Bayesian Regularization(BR)SOM clustering algorithm was established to analyze the power consumption behavior model of the distribution network.There are closely related nodes and neighborhood nodes in the extension of the output layer,which can learn according to their own characteristics.Therefore,adjacent nodes will form similar weights and achieve good matching with similar input nodes.After that,a power grid in Ningxia was selected as the test object to verify the effectiveness of this method.The research results show that the improved Relief algorithm is used to screen the power system load feature vectors,and finally 5 feature vectors are determined,which greatly reduces the number of feature variables and achieves the effect of simplifying load description.The average accuracy of three types of user loads with significant characteristics is tested,and the accuracy of SOM algorithm is above 97%,which verifies the accuracy of the model in this paper.