A novel method for local anomaly detection of time series based on multi entropy fusion

Abstract

Dynamic Shannon entropy, which is fused by Shannon entropy and permuted distribution entropy (PDE), achieves good results in local anomaly recognition of time series. However, the dynamic Shannon entropy performs poorly in identifying various abnormal behaviors of complex signals with noise and chaotic characteristics and periodic background. It also gets unsatisfying results in anomaly detection of signals containing relatively large uncertainty anomaly information. Deng entropy, an extended entropy of Shannon entropy, expands the information capacity of Shannon entropy and is more excellent at describing uncertainty and complexity. In this paper, dynamic Deng entropy (DyEd) is proposed by fusing Deng entropy and PDE to identify local anomalies of time series. Six numerical experiments and an empirical application are conducted to illustrate the efficiency of the proposed method. Results show that the proposed method is of great success for signal anomaly detection with periodic backgrounds.