基于特征加权的SFO-K_means用电行为研究

Research on electricity consumption behavior based on feature weighted SFO-K_means

在传统的K_means算法中,初始聚类中心大多采用任意选取或者凭借经验,使算法的准确性易受选取结果的影响。针对这些不足,提出了利用剑鱼算法全局寻优特性,来改进K_means初始聚类中心的选取。UCI数据对比显示,改进算法在平均迭代次数和准确率方面优于传统K_means算法。在用户用电行为分析上,利用提取的5个降维特征指标,对每半小时采样一次的高维日负荷曲线进行降维处理;引入Critic法来确定指标的权重系数,并利用皮尔逊相关系数与熵权,分别衡量指标之间的冲突性与指标内部的对比强度;采用特征加权的SFO-K_means算法进行聚类分析。算例结果表明,该法在聚类质量上有一定的优越性,聚类结果能反应出用户的用电行为习惯。

In the traditional K_means algorithm,the initial clustering centers are mostly selected arbitrarily or with experience,and the accuracy of the algorithm is easily affected by the selection results. In view of these shortcomings,this paper proposes to improve the selection of K_means initial clustering centers by using the global optimization characteristics of the Sailfish Optimizer(SFO). The comparison of UCI data shows that the improved algorithm is better than the traditional K_means algorithm in the average number of iterations and accuracy. In the analysis of users’ electricity consumption behavior,firstly,the extracted five dimensionality reduction characteristic indexes are used to reduce the dimensionality of the high-dimensional daily load curve which is sampled every half hour.Then the Critic method is introduced to determine the weight coefficient of the indexes,and the Pearson correlation coefficient and entropy weight are used to measure the conflict between the indexes and the contrast strength within the indexes.