Real-time anomaly detection in gas sensor streaming data

Abstract

In order to improve the timeliness and accuracy of coal mine gas disaster risk assessments, it is important to detect anomalies in the gas sensor streaming data in real-time. In this paper, the support vector regression (SVR) algorithm combined with the normal statistical distribution technique is used to establish a real-time anomaly detection model for gas sensor streaming data. Furthermore, a prototype system for the real-time anomaly detection in gas sensor streaming data that is built using the stream processing framework Spark Streaming is presented. Experiments show that the real-time anomaly detection system can periodically update the anomaly detection model and determine anomalies in the sensor streaming data in real-time. For a window size of 9, an update cycle of 1 and an anomaly threshold of 0.95, the anomaly detection model is better than the boxplot, the statistical analysis and the clustering algorithm regarding the prediction precision and accuracy.