A power time series clustering method based on maximum spanning tree and anomaly detection

CN122087282APending Publication Date: 2026-05-26SHANGHAI YIJUNENG ENERGY INFORMATION TECHNOLOGY CO LTD

Patent Information

Authority / Receiving Office: CN · China
Patent Type: Applications(China)
Current Assignee / Owner: SHANGHAI YIJUNENG ENERGY INFORMATION TECHNOLOGY CO LTD
Filing Date: 2026-02-10
Publication Date: 2026-05-26

AI Technical Summary

Technical Problem

Existing power time series clustering methods are sensitive to scale variations and noise in the time series, making it difficult to characterize complex fluctuation and nonlinear features. Furthermore, they rely on manual selection of distance metrics and the number of clusters, leading to unstable clustering results.

Method used

A global similarity structure for power time series is constructed based on the maximum spanning tree, and anomaly detection is used for pruning to adaptively form a clustering structure, avoiding manual parameter selection. The maximum spanning tree is constructed using similarity coefficients, and outlier branches are identified and pruned by anomaly detection, achieving clustering without the need to preset the number of categories.

Benefits of technology

Adaptive power time series clustering is achieved, which improves the stability and rationality of clustering results, adapts to multi-mode and highly fluctuating power data, provides clustering results with strong structural interpretability, and serves predictive modeling and scheduling decisions.

✦ Generated by Eureka AI based on patent content.

Smart Images

Figure CN122087282A_ABST

Patent Text Reader

Abstract

This invention relates to a power time series clustering method based on maximum spanning tree and anomaly detection. The method includes: acquiring a first power time series to be clustered; preprocessing it to obtain a second power time series; calculating similarity coefficients; constructing a similarity fully connected graph by treating each second power time series as a node and using any one or the average of multiple similarity coefficients between any two nodes as the distance weight between those nodes; constructing a maximum spanning tree based on the similarity fully connected graph; performing anomaly detection to identify anomalous branches with short lengths that exhibit outlier characteristics in the overall branch length distribution, and pruning these anomalous branches to divide the maximum spanning tree into multiple subtrees, with each subtree containing a set of time series as a cluster category. This invention effectively avoids the problem of traditional clustering methods being highly sensitive to parameter selection.

Need to check novelty before this filing date? Find Prior Art