基于改进密度聚类的用气异常检测

Anomaly Detection of Industrial Gas Load Based on Improved Density Clustering

为了更好地从管网系统中挖掘数据信息,科学地计量工商业用气规律,帮助燃气公司对用户异常用气行为进行智能识别,本文提出了一种基于K-近邻距离图和网格搜索法(Grid search)的密度聚类(DBCSAN)算法,结合分段聚合近似表示方法(PAA)在包含噪声的数据集中通过寻找工业燃气数据集的内在分布规律和聚类效果的变化来识别异常点。首先以来自SCADA和智能表具采集的南方某陶瓷工厂日负荷数据为例,使用PAA方法对数据进行降维处理。其次利用改进的DBSCAN算法对案例用户监测时段中的异常数据进行识别。最后将算法在某南方陶瓷行业的325个用户数据上进行了验证。结果表明,算法的平均准确率在90%以上,人工智能算法在燃气领域的应用对于燃气经营企业实现精细化管理、以及达到降本增效的效果具有一定的指导意义。

In order to better mine the data information from the pipeline network system, scientific measurement of industrial and commercial gas consumption laws, and help gas companies to intelligently identify abnormal gas consumption behavior of users, a density clustering (DBCSAN) algorithm based on k-nearest neighbor distance graph and Grid search is proposed, which identifies outliers in noisy data sets by the method of approximate representation based on piecewise aggregation (PAA) to search for the distribution of industrial gas data sets and the change of clustering effect. Firstly, the daily load data of a ceramic factory in south China collected by SCADA and smart meters are taken as an example, and the data are processed with PAA method. Secondly, the improved DBSCAN algorithm is used to identify the abnormal data in the monitoring period of case users. Finally, the algorithm is applied to 325 users’ data in a ceramic industry in south China. The results show that the average accuracy of the algorithm is over 90%, and the effective application of artificial intelligence algorithm in gas field is realized. It has certain guiding significance for gas enterprises to realize fine management and achieve the effect of reducing cost and increasing efficiency.