To address the challenges of timely identification of abnormal market participants before electricity spot market clearing and quantitative attribution of the underlying causes of abnormal behaviors

a market power behavior identification and attribution method for intra-market supervision is proposed. First

a market power indicator system covering physical and economic behaviors and an indicator-behavior mapping relationship are established. Second

an isolation forest is used for unsupervised detection of multidimensional features

including bidding

quantity declaration

maintenance declaration

and energy storage state of charge

thereby obtaining anomaly scores and abnormal states for market participants. Then

a causal forest estimates the conditional average treatment effect of anomalous states on market outcomes

while CatBoost extracts the explanatory weights of individual indicators for the heterogeneity of these effects. Finally

typical market power behaviors are attributed through the indicator-behavior mapping relationship. Case studies using actual data from a regional electricity market in China show that the model can accurately identify various abnormal patterns

including physical withholding

economic withholding

and collusive behaviors

achieving high precision and recall. The proposed method can provide a basis for identifying abnormal market participants and analyzing the underlying causes of their behaviors for intra-market supervision of electricity spot markets.