Clustering analysis of load curve based on time series density-changing processing

Abstract

Load curve clustering constitutes the foundational methodology for the analysis of customer load characteristics, enabling the extraction of typical power consumption patterns from a substantial volume of load data. Understanding these patterns is pivotal for applications including demand response, tariff design and power grid planning. In light of the inadequacies of existing clustering methodologies, specifically their insufficient consideration of the unique characteristics of load time periods, an advanced clustering technique based on time series density-changing processing is introduced in this study. The proposed method commences with the application of linear interpolation to augment the density of data points during critical periods, including peak, trough and ramp-up periods, thereby accentuating and amplifying their influence within the clustering framework. Concurrently, the technique incorporates the adaptive piecewise aggregate approximation (APPA) for dimensionality reduction to mitigate the density of superfluous data. Subsequently, a comprehensive index, formulated through the integration of Euclidean distance and correlation distance, is employed to conduct k-medoids clustering analysis of the load curves. The effectiveness of the proposed methodology is validated utilizing actual residential customer data from the UCI dataset. The results of these empirical investigations affirm that the method significantly enhances the efficacy of load clustering, thereby providing an authentic representation of the power consumption characteristics of residential customers.

Keywords

• density-changing
• load clustering
• critical periods
• ordinary periods
• interpolation
• adaptive piecewise aggregate approximation (appa)