Real-time anomaly detection and analysis of time series data for crack gauge in landslides
In order to address the issue of effectively identifying abnormal data in real-time monitoring of landslide cracks,occasional abnormal thresholds based on the various deformation stages of landslides are established.Additionally,a real-time anomaly detection method for time series data,utilizing interval prediction,is proposed.This method takes into consideration the temporal logic relationship between data and the correlated information of landslide deformation stages.Firstly,the time series characteristics of cumulative displacement of crack gauge are extracted using the autoregressive integrated moving average(ARIMA)model.Subsequently,an interval prediction model is constructed,and the sliding window algorithm is employed to predict sub-sequences.Secondly,to determine prospective abnormal points,a modified confidence interval(with a=0.05)is utilized,and occasional abnormal thresholds are established for different deformation stages of landslides.Finally,exceptional information is obtained through combined anomaly recognition.The research results indicate that this method accurately identifies abnormal data values and demonstrates universal applicability in real-time anomaly detection of time series data.By comparing the predicted interval of the model with the abnormal values,the real-time possibility of data abnormalities can be obtained.Furthermore,this method provides valuable data reference for intelligent decision-making in landslide monitoring and early warning.
slope engineeringlandslidetime series modelreal-time anomaly detectionmonitoring and early warning
万方数据
维普
