Anomaly Detection Approach for Sensor Networks in Coal Mine Solid Backfilling Working Faces Based on Transformer

In this paper, we propose a comprehensive anomaly detection model for the perception system of backfilling working faces. The model consists of three main components: the position encoding module, the time series prediction module, and the anomaly detection module. The position encoding module extracts global temporal information, including year, month, and day, based on the timestamp information of the data. The time series prediction module utilizes the Temporal Fusion Transformers model, which considers the latent correlations among different variables for accurate predictions. The anomaly detection module assesses the data's abnormality by comparing the predicted values with the actual values. We validate the performance of the model on a real-world dataset, demonstrating its high accuracy and strong generalization capabilities for various types of sensor data in backfilling working faces. This research provides a viable solution for anomaly monitoring and management in backfilling operations, with significant practical implications.