Application of machine learning in the monitoring of power market collusion behavior

With the global progress of power market reform, especially in China where policies have spurred an influx of diverse market players, both the number of market entities and the scale of market transactions have grown significantly. However, this development has given rise to credit risk issues, among which collusion has become increasingly prominent. Collusion, manifested in forms like price manipulation by power generation companies during peak - demand periods and coordinated bidding, seriously endangers the fairness and stability of the power market, leading to inefficiencies in resource allocation and harm to the long - term development of the power industry.To address this, this paper focuses on monitoring indicators and early - warning methods for collusion in the power market. It innovatively constructs an unsupervised machine learning model based on graph convolutional clustering. By treating market entities as nodes and their relationships as edges in a graph, the model collects extensive data and extracts complex features through graph convolutional operations. It then clusters companies according to these features, identifying potential collusion when companies show similar abnormal behavior patterns. The model also quantifies the suspicion level of each company or group based on multiple factors. This research offers robust technical support for power market regulation, enabling real - time market monitoring, timely regulatory intervention, and ultimately contributing to the healthy operation of the power market, fair competition, and sustainable development of the power industry.