Dynamic real–time abnormal energy consumption detection and energy efficiency optimization analysis considering uncertainty

Abstract

Uncertain working conditions will lead to abnormal energy consumption and energy loss of high energy consumption machines. Therefore, it is necessary to detect abnormal energy consumption data, which is designed to optimize energy efficiency. In this study, two unsupervised anomaly detection methods are proposed, suitable for dynamic real–time abnormal energy consumption detection. As for point anomaly, a Rain Flow–based Connectivity Outlier Factor algorithm is proposed, which is a connectivity–based detection method. While, as for collective anomaly, a Rain Flow–based Mean Nearest Neighbor Distance Anomaly Factor algorithm is proposed, which is a distance–based detection method. The proposed two algorithms adopt the methods of time sorting and piecewise cubic Hermite interpolating polynomial, respectively. The aim is to accurately locate anomalies in the time dimension. Then, abnormal energy consumption is automatically found by setting a dynamic real–time threshold. To verify the superiority and the effectiveness of the proposed method, the actual energy consumption data from aluminum extrusion enterprises are used for simulation experiments. The results demonstrate that the proposed method can detect anomalies quickly and accurately. Finally, this study presents an energy efficiency optimization framework based on anomaly detection. Our tests in energy management systems show that the proposed framework can optimize energy efficiency and reduce energy waste. This provides an effective solution to clean production and low carbon manufacturing of high energy consumption machines.