An investigation into detecting anomalous trading patterns in electricity markets utilizing a SMOTE-CMAES-LightGBM model

In response to the increasingly prominent issue of regulatory violations in electricity market trading, efficiently identifying the categories of misconduct committed by market participants has become a key research focus. To address this challenge, this study proposes an intelligent identification model to detect multiple types of anomalous trading behaviors, including collusion, free-riding, and abnormal price bidding in electricity markets. The study first uses gray relational analysis (GRA) to screen strongly correlated monitoring indicators and construct a feature set. It then introduces the covariance matrix adaptive optimization strategy (CMAES) to optimize the hyperparameters of the lightweight gradient boosting machine (LightGBM) to improve tuning efficiency and reduce manual reliance. The CMAES-LightGBM hybrid model constructed on this basis uses the synthetic minority oversampling technique (SMOTE) to effectively alleviate data imbalance and achieve efficient identification of multiple types of illegal behaviors. Experimental results show that the model achieves an 92.6% recognition accuracy and an average computational time of 33 milliseconds, validating its accuracy and timeliness in power market applications.