Anomaly Detection For Safety Monitoring Data Of The South-to-north Water Transfer Project Based On Time Series Mining Technology

Time series mining is formed on the basis of traditional sequential analysis and it is one of the most challenging research directions in data mining.In various fields,time series mining technology has been widely used.However,different application background has different practical requirements for mining technology,it still needs to study sequential mining algorithm and model constantly.As one of time series mining,abnormal detection is an important technology in many areas,and it raises more and more widespread attention,becoming a hot research issue.The middle route project of south-to-north water diversion project is a strategic project to alleviate the shortage of water resources in northern China and optimize the allocation of water resources in our country.The complicated geological conditions and meteorological conditions along the middle route of the south-to-north water diversion project pose severe challenges to the safety of the project.In order to ensure the safety of the project,a large number of sensors have been laid along the route of the south-to-north water diversion project to collect real-time project safety data.There are still some problems in the analysis of the safety monitoring data of the middle route project of south-to-north water diversion project.The complexity of sensor working environment leads to abnormal safety monitoring data is a problem.To ensure the safety of the middle line of south-to-north water diversion project,this paper studies the abnormal detection of monitoring data,the main research work are as follows:First,by introducing the dynamic time bending method,the abnormal detection of the segmented monitoring data is completed.the algorithm first uses piecewise linear representation based on key points to segment the original time series into a pattern of unequal length,and then uses dynamic time bending method to detect the anomaly of the segmented pattern.The feasibility and validity of the algorithm are verified by experiments on the safety monitoring data of the middle route project of south-to-north water transfer project.Second,the cross-correlation analysis of time series fractal analysis is used to analyze the correlation of safety monitoring data of the middle route project of southto-north water diversion project.cross correlation analysis is applied,such as Pearson coefficient method,Sparkman Rank correlation coefficient method and trend cross correlation coefficient method are applied to the safety monitoring data of the middle route project of south-to-north water transfer project respectively to conduct the correlation analysis of different sensor data.The sensor sequences with high positive correlation are obtained.Based on the anomaly detection results and the highly positive correlation of sensor sequences,simple judgment of the anomaly of multi-correlation sensor data is realized.Third,the sensor data with high positive correlation are detected by random forest algorithm,and the cross correlation coefficient method was used to calculate the correlation coefficient between monitoring data,through experiments and comparison of the results,the anomaly was detected accurately and the validity of the algorithm was verified.