Application Of Imbalanced Data Clustering In Abnormal Detection Of Building Energy Consumption

Imbalanced data means that the number of samples contained in different classes of data sets is greatly different or the number of samples in different classes is the same but the distribution of these samples is uneven.In the abnormal detection of building energy consumption,there are great differences in the number and distribution between normal energy consumption data and abnormal energy consumption data.In essence,abnormal detection of energy consumption belongs to the problem of data imbalance.For imbalanced data,traditional clustering methods tend to equalize the number of samples in each class,resulting in a high rate of misclassification.In order to avoid this "uniform effect",this paper studies the problem of imbalanced data clustering,and applies the innovative results to the abnormal detection of building energy consumption.The specific contents are as follows:(1)Aiming at the data overlapping problem in the data set,a generalized fuzzy c-means(GFCM)clustering algorithm is proposed under the framework of D-S evidence theory.By setting the meta-cluster threshold,the membership matrix of traditional FCM is extended,and the samples that are difficult to be divided into a specific cluster are divided into meta-cluster according to the obtained generalized membership matrix.The algorithm represents the overlapped(uncertain)samples reasonably,and reduces the risk of misclassification.(2)Aiming at the "uniform effect" of traditional clustering algorithm on imbalanced data clustering,a multi-partition(MP)clustering algorithm for imbalanced data based on D-S theory is proposed,which clusters imbalanced data through four sub-steps: multi-partitioning of the data set,searching for the number of real clusters,merging sub-data set and partitioning with the remaining samples.And it reasonably represents the overlapping samples in data set,reduces the error rate and avoids the "uniform effect" of samples.Compared with other clustering algorithms,the effectiveness of MP algorithm is verified.(3)MP clustering algorithm is applied to the abnormal detection of building energy consumption.By collecting and preprocessing the building energy consumption data of a large-scale shopping mall in Xi'an,and clustering the processed data with MP algorithm,the abnormal value of energy consumption is obtained.The effectiveness of MP algorithm is further verified by the analysis of specific experimental results of abnormal detection of energy consumption.