基于多判据融合的用电信息采集系统异常数据甄别模型

Abnormal Data Discrimination Model for Electricity Information Acquisition System Based on Multi-criteria Fusion Method

针对海量数据处理难度大,传统人工排查方法效率低、实时性不高等问题,建立了基于多判据融合的用电信息采集系统异常数据甄别模型。首先,对用电信息采集系统数据断点、异常点和现场实际运行数据情况进行统计分析;分别采用原型聚类法、密度聚类法、概率密度法和深度学习方法4种方法进行异常值甄别,并比较各方法的异常值甄别结果;为避免单一判断准则的随机性与不准确性,将4种方法异常值甄别结果进行相互交叉验证,获得最终的异常值甄别结果;基于已经训练完成的模型,在线监测异常数据,最终建立基于多判据融合的异常数据甄别模型。通过用电信息采集系统异常数据甄别模型基本测试和实际电表的电流和功率测试,结果验证了模型和方法的可行性和有效性。

To solve problems of difficulty in massive data processing and low efficiency and low real-time of traditional artificial screening method,an abnormal data discrimination model for the electricity information acquisition system based on multi-criteria fusion.Firstly,this paper statistically analyzes data breakpoints,outliers and actual operating data of the electricity information acquisition system,and then uses four methods including the prototype clustering method,the density clustering method,the probability density method and the deep learning method for outliers discrimination and compares discrimination results of different models.To avoid randomness and inaccuracy of single criterion,the discrimination results of outliers of four methods are cross-validated to obtain finally discrimination results.On the basis of the trained models,abnormal data are monitored online,and the abnormal data discrimination model based on multi-criteria fusion is finally established.Basic test of the abnormal data discrimination model for the electricity information acquisition system and current and power test of the actual ammeter are performed to verify feasibility and effectiveness of the proposed model and method.