MetaFormer-PAD: A Novel Meta-learning Framework for Anomaly Detection in Power Trading

In power markets, the scarcity of anomaly samples and the continuous changes in market rules and trading patterns create challenges for anomaly detection. To tackle this, the paper proposes the MetaFormer-PAD model to address the challenges of label scarcity and dynamic patterns in power trading anomaly detection. MetaFormer-PAD integrates meta-learning and few-shot learning techniques, enabling the model to quickly adapt to new anomaly patterns with limited labeled data, enhancing its robustness and performance in dynamic market conditions. Experimental results show that MetaFormer-PAD outperforms existing methods on the PJM and ELECON datasets, particularly in detection speed and anomaly diversity recognition, proving its effectiveness in detecting anomalies in power markets where data is scarce and conditions change rapidly.