Analysis of Behavioral Characteristics of Trading Entities in the Electricity Market Based on a Time Series Clustering Method

This study introduces an innovative time series clustering approach to analyze the behavior of trading entities in the electricity market. The method incorporates wavelet transforms for noise attenuation, auto-encoders for deep feature extraction and DTW k-means clustering to overcome the limitations of traditional clustering methods, which tend to ignore the frequency specificities of time series data. Using transaction data from the provincial power market, we construct an indicator system and apply our proposed method to reveal distinct consumption patterns and preferences among different entity types. The empirical results show that our approach is significantly better than conventional methods in terms of clustering accuracy and robustness as measured by the silhouette coefficient and the Davis-Bouldin index. This study provides valuable insights for the optimization of electricity trading and a new analytical tool for understanding the complex dynamics of electricity market behavior.