Detection algorithm for anomaly sub-sequences in energy consumption data stream
Zhihui Ye, Jie Qian, Wenjuan Wang, Xiaohu Hu · International Conference on Mechanisms and Robotics (ICMAR 2022) · 2022
For the data streams retrieved from the various sensors in energy cyber physical system, an anomaly detection algorithm based on time series extremum was proposed. Firstly, for a single data stream, the algorithm identifies the extreme points of the data stream in the sliding window, and marks off the abnormal sub-sequences according to these points. Then the eigenvalues of the sub-sequences are calculated to establish the feature space. Several anomaly measures are defined, so the sub-sequences can be sorted by the measures in the feature space. Finally, based on the correlation between the sensors points, the anomaly points on each work section are detected. The algorithm can make up for the limitation of other point anomaly detection algorithms by detecting the abnormal sub-sequences, and reduces the probability of false alarm and missing alarm in data stream analysis. Experiments show that the algorithm is effective and feasible.