With the increasing intelligence of the power grid and the deepening reform of the power market, the risk that some market participants violate trading rules and engage in bidding collusion has become increasingly prominent. These risks pose new challenges to the stable operation of the power grid and the power market. Meanwhile, identifying potential collusion patterns in real-world market scenarios remains extremely challenging. To tackle these issues, we propose Q UICK- O PTICS Clustering with Bid ding Data-Driven Indicators for C ol lu sion D e tection ( QO-BIDCLUE ), a novel method for collusion identification in centralized power market bidding using bidding data-driven indicators. First, we construct a multidimensional indicator system based on bidding data to quantify the bidding interactions among market participants. Second, we develop a density clustering model based on Quick Ordering Points to Identify the Clustering Structure, which can autonomously mine behavioral patterns without labeled training data. Finally, we conduct a detailed case study using the bidding data of power sales companies from a specific province in China. Quantitative and qualitative experimental results demonstrate that our method can efficiently and accurately detect collusive market entities, thus validating its effectiveness. The QO-BIDCLUE method can further generate a watchlist of suspected collusive entities for regulators from massive power market transaction data. Overall, this work provides a data-driven tool to reinforce market supervision and sustain the coordinated stability of power grid operation and power market operation. • We propose a multi-dimensional indicator system driven by three-segment bidding data. • We adopt an unsupervised density clustering method to identify suspected collusion. • Empirical results from real market bidding data verify the method’s effectiveness.
Li et al. (Sun,) studied this question.