Abnormal detection of electric power marketing data based on BiLSTM-CNN-CRF

Power marketing data exhibits clear time series characteristics, and anomalies are often closely related to equipment failures, fraudulent activities, or sudden changes in consumer behavior. This paper proposes an anomaly detection method for electric power marketing data based on a BiLSTM-Convolutional Neural Network (Convolutional Neural Networks, CNN)-Conditional Random Field (Conditional Random Field, CRF) model, aiming to accurately identify abnormal behaviors through in-depth analysis of historical sales data. Using sales data from the State Grid as the experimental subject, the data was first preprocessed, including normalization and filtering. A BiLSTM-CNN-CRF model was then constructed and trained. By capturing long-term dependencies in the data, the model extracted features from the time series, effectively identifying anomalies in power marketing data. Several metrics were selected in the experimental design to evaluate the model's prediction accuracy, ensuring the results demonstrate that the anomaly detection model based on BiLSTM-CNN-CRF exhibits high accuracy and stability in identifying anomalies in power marketing data. The model performs well on test data, effectively predicting future anomalies, and surpasses traditional methods in both efficiency and precision. This research provides support for power companies to achieve intelligent data analysis in a complex market environment, contributing to the improvement of resource allocation efficiency and risk prevention capabilities.