基于双尺度度量的改进模糊均值曲线聚类方法研究

Research on improved fuzzy mean curve clustering method based on two-scale measurement

智能化电力网络中存在许多随时间变化表现出明显曲线特征的函数型数据,进行曲线聚类可以有效挖掘数据信息。针对模糊均值聚类算法初始聚类中心选取困难以及曲线聚类方法中相似度衡量不准确等问题,提出一种基于双尺度度量的改进模糊均值曲线聚类方法。根据皮尔逊距离衡量曲线的纵向形状相似性、动态时间弯曲距离衡量曲线的横向形状相似性,提出一种基于双尺度度量的密度峰值算法确定初始聚类中心;采用改进熵权法融合皮尔逊距离与动态时间弯曲距离作为聚类算法中的相似性度量;采用聚类有效性指标,从聚类效果、算法稳定性2个方面对聚类结果和算法性能进行评价;最后采用某地区一年的风电实际出力数据作为算例样本进行聚类分析,验证了所提出模型及算法的正确性和有效性。

There are many functional data showing obvious curve features that vary with time in intelligent power networks.Curve clustering can effectively mine the data information.Aiming at the difficulty in selecting the initial clustering centre for fuzzy mean clustering algorithm and the inaccurate similarity measurement of curve clustering methods,an improved fuzzy mean curve clustering method based on two-scale metric is proposed.The longitudinal shape similarity of a curve is measured according to the Pearson distance,and the horizontal shape similarity of the curve is measured according to the dynamic time wrapping distance.Then,a density peak algorithm based on two-scale measurement is proposed to determine the initial clustering centre.The improved entropy weight method combines Pearson distance and dynamic time wrapping distance in similarity measurement of clustering algorithm.Clustering validity indexes are taken to evaluate the clustering results and algorithm performance from the aspects of clustering effect and algorithm stability.At last,taking the annual data of wind power outputs in a region as the example for clustering analysis,the results verify the correctness and effectiveness of the model and calculation method.