Abnormal Detection Method Of Gas Heating Energy Consumption

With the popularization of gas centralized heating and the increase of the number of gas boilers,the structure of gas heating system becomes more and more complex,which leads to different degrees of potential safety problems.Abnormal detection of gas heating energy has become a key area of safety control.Gas heating data has high dimension,few labels,unknown anomaly ratio,complex data components,and the accuracy of anomaly detection algorithm is easily affected by the anomaly ratio parameters.Therefore,traditional detection methods are not applicable to this scene.There are a large number of false or missed detection in the actual detection,which greatly reduces the safe operation of heating energy equipment.Therefore,in order to reduce the rate of false detection and false detection of gas heating energy,two kinds of anomaly detection algorithms for gas heating energy are proposed.The main research contents are as follows:First,an anomaly detection algorithm based on iForest-DEC is proposed.The algorithm first uses the iForest(Isolation Forest)algorithm with high anomaly ratio to perform pre-anomaly detection,and then uses the DEC(Deep Embedded Cluster)algorithm to perform secondary fine detection.The pre-anomaly samples extracted by the iForest algorithm are subdivided,which solves the problem that the anomaly ratio setting affects the performance of the algorithm and improves the accuracy of anomaly detection.The experiment shows that iForest-DEC algorithm can effectively get rid of the influence of abnormal ratio parameters on detection accuracy,and has good performance in gas heating scenarios.The average A UC value of this algorithm on three sets of gas heating data sets is 23.57%higher than iForest algorithm.Secondly,an anomaly detection algorithm based on iForest-LOFDEGDC is proposed.In order to ensure the accuracy of anomaly detection and further reduce the rate of missed detection,dissertation makes two improvements to the iForest-DEC algorithm:(1)The iForest-LOF fusion algorithm is proposed to replace the iForest algorithm for pre-anomaly detection,so as to take into account the global information and local information and reduce the missed detection rate of anomaly;(2)The improved DEGDC(Deep Embedded Grid Density Clustering)algorithm based on DEC is proposed for secondary fine detection.The lowdimensional feature extraction is carried out through SSAE(Stacked Sparse Auto-Encoder)network,and the grid density is introduced to optimize the initial clustering center to improve the clustering accuracy and anomaly detection accuracy.The experiment shows that the iForest-LOFDEGDC algorithm proposed in dissertation has high Precision and Recall rate in three groups of gas heating data sets and two groups of open data sets.The maximum AUC value of the algorithm in the gas heating energy consumption scenario can reach 0.974,and the maximum Recall rate can reach 0.994,which can effectively reduce the abnormal missing detection rate and false detection rate,and has significant advantages.