Research on the data validity of a coal mine solid backfill working face sensing system based on an improved transformer

Abstract

Abstract Solid backfilling in coal mining refers to filling the goaf with solid materials to form a support structure, ensuring safety in the ground and upper mining areas. This mining method maximizes coal production and addresses environmental requirements. However, in traditional backfill mining, challenges exist, such as limited perception variables, independent sensing devices, insufficient sensing data, and data isolation. These issues hinder the real-time monitoring of backfilling operations and limit intelligent process development. This paper proposes a perception network framework specifically designed for key data in solid backfilling operations to address these challenges. Specifically, it analyses critical perception objects in the backfilling process and proposes a perception network and functional framework for the coal mine backfilling Internet of Things (IoT). These frameworks facilitate rapidly concentrating key perception data into a unified data centre. Subsequently, the paper investigates the assurance of data validity in the perception system of the solid backfilling operation within this framework. Specifically, it considers potential data anomalies that may arise from the rapid data concentration in the perception network. To mitigate this issue, a transformer-based anomaly detection model is proposed, which filters out data that does not reflect the true state of perception objects in solid backfilling operations. Finally, experimental design and validation are conducted. The experimental results demonstrate that the proposed anomaly detection model achieves an accuracy of 90%, indicating its effective detection capability. Moreover, the model exhibits good generalization ability, making it suitable for monitoring data validity in scenarios involving increased perception objects in solid backfilling perception systems.