基于数据挖掘的建筑能耗异常检测研究

Research on Abnormal Detection of Building Energy Consumption Based on Data Mining

摘要:

建筑能耗异常检测对于建筑管理和运行至关重要,论文提出了一种基于D-S证据理论的不平衡数据多划分(Multi-partition,MP)聚类算法,并构建MP算法能耗异常检测模型对建筑能耗中的异常值进行准确检测。首先通过改进的信任c均值算法将能耗数据集多划分；利用基于K-NN的均值漂移算法确定数据集的真实类别个数；然后根据密度合并规则对能耗数据进行合并；最后对未合并的能耗数据再次划分得到最终的能耗异常检测结果。UCI数据集验证结果表明,MP算法对于不平衡数据聚类效果良好,能够有效避免样本“均匀效应”,降低错误率；通过对某大型商场建筑空调和照明用电能耗异常值检测,验证了MP算法能耗异常检测模型的有效性。

Abstract:

Abnormal detection of building energy consumption is very important for building management and operation. In this paper, a multi-partition (MP) clustering algorithm for imbalanced data based on D-S evidence theory is proposed, and the energy consumption anomaly detection model of MP algorithm is constructed to accurately detect the abnormal values in building energy consumption. Firstly, the energy consumption data set is divided into multiple parts by the improved credal c-means algorithm. The KNN-based Mean-shift algorithm is used to determine the number of real categories of the data set. Then the energy consumption data is merged according to the density merging rules. Finally, the energy consumption data that is not merged is divided again to get the final abnormal energy consumption detection results. The UCI data set verification results show that the MP algorithm has a good clustering effect for imbalanced data, which can effectively avoid the "uniform effect" of samples and reduce the error rate. Through detecting the abnormal values of the energy consumption of air conditioning and lighting in a large shopping mall, the validity of the energy consumption anomaly detection model of MP algorithm is verified.