Tweet Share

• Send by email

  Send by email

  

  Online Multidimensional Time Series Anomaly Detection Algorithm based on Shield Construction

  The real-time anomaly detection is very important in the process of shield tunneling. The alarm for abnormal events can help the staff to make decisions quickly and ensure the engineering safety. However, the traditional methods of shield anomaly detection are mainly depended on analysing for one-dimensional time series one by one, so its accuracy is low, which cannot meet the actual engineering needs. Therefore, this paper proposes RADELW (Real-time Anomaly Detection based on the Distance of Extended Frobenius Norm with Local Weight) to solve this problem by joint analysis of multi-dimensional time series. An Extended Frobenius Norm Distance with local weight is designed and is used to difference of two multidimensional time sub-sequence, and the anomaly detection is based on first order difference of the distance in real-time. In this paper, RADELW is applied in Shanghai and Hangzhou tunnel engineering respectively. The experimental results show it has high accuracy and a low false alarm rate, so it is proved to be feasibility for shield tunneling anomaly detection.

  Recipient :

  Name ^* :

  E-mail address ^* :

  ^* Required fields

  Send   or  Cancel

• Print