1. An innovative dynamic anomaly detection method based on hybrid data mining technology for building energy consumption.
- Author
-
Lei, Lei, Guo, Xuesong, and Zheng, Run
- Abstract
Energy waste caused by equipment malfunction or inefficient energy utilization is common during the building operation process. Through the anomaly detection of building energy consumption data, the utilization rate of building energy can be improved. A dynamic anomaly detection method for building energy consumption data point anomalies and collective anomalies is proposed in this paper. In the method, the entropy-weighted k-means (EWKM) is coupled with the classification and regression tree (CART) to establish a recognition and matching mechanism for unlabeled energy consumption data. Then, the isolated forest (IF) and local outlier factor (LOF) are combined to enhance the detection ability of point anomaly. Through dynamic time warping, the distance between energy consumption sequences is calculated to realize the collective anomaly detection of energy consumption. The historical dataset could be dynamically updated based on the detection results. Subsequently, the dynamic anomaly detection of EWKM-CART-IF-LOF is achieved. The proposed EWKM-CART-IF-LOF is applied to detect point and collective anomalies in the real energy consumption of an office building in Suzhou. In the test set, the detection performances of EWKM-CART-IF-LOF, EWKM-CART-IF, EWKM-CART-LOF, EWKM-CART-CBLOF and EWKM-CART-KNN are evaluated and compared. Results show that the EWKM-CART-IF-LOF has the best detection performance with an AUC of 0.90, an accuracy of 0.98, a precision of 0.83, and an F-score of 0.82. These outcomes validate the method's efficacy in optimizing building energy consumption, making a significant contribution to the fields of energy management and anomaly detection. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF