A new fusion model for anomaly detection of gas data

Abstract

Accurate gas data are needed to support the construction of gas energy consumption monitoring system. In the sampling process, we found that some gas data had certain errors. In order to improve the accuracy and reliability of gas data, this paper deeply studied and analyzed the advantages and disadvantages of four algorithm models, k-means, LOF, isolated forest and One-Class SVM.A fusion algorithm model based on the above four models is proposed to realize the multi-dimensional complementarity of the four basic models. Anomaly detection is carried out on two groups of gas data at two sampling points by using this model to find abnormal points to improve the data quality. Finally, the experiment proves that the fusion algorithm improves the accuracy of detection, saves the running time, and achieves satisfactory results in gas data processing.