Research On Identification And Evaluation Method Of Disaster Risk Segment Based On IMU Strain Data

Long distance oil and gas pipeline will bend deformation and pipe position shift when affected by geological disaster,which will lead to safety accidents and endanger the safety of people’s lives and property.Periodical inspection of pipeline safety conditions can effectively reduce the occurrence of such accidents.The bending strain data of the whole line of the pipeline can be detected by using the inertial measure unit(IMU).Based on the big data of IMU strain detection of a certain section of frozen soil for many years,an intelligent identification and evaluation method of geological hazard based on IMU strain data is proposed for the difference of the strain curve characteristics between the geometric characteristics and the bending deformation section of the pipeline.In this thesis,attitude data detected in IMU are firstly converted into strain data,and then four main types of locally deformed pipe sections in the pipeline are summarized and classified.The interference noise in IMU strain data is removed by wavelet decomposition method,and more than 30,000 samples is extracted from IMU strain data from a frozen region for a total of five years from 2014 to 2018.Accroding to the pipeline testing information such as location,date of test sample database is established,from the sample data to extract the time and frequency domain characteristics of a total of 36,retained by the principal component analysis before 9 principal components to complete the feature data dimension reduction,constructs the decision tree and support vector machine(SVM),random forests,such as machine learning classification model completed the classification of the recognition of different local deformation section,Girth weld information construct sample data at the same time,A one-dimensional convolutional neural network deep learning model is established to complete the classification,comparing the data of different secondary strains,the absolute change of strain is calculated,and the pipe sections with large bending strain changes are identified,comprehensive evaluation based on the identification results of geological hazard segments,finally combining GPS data visualization of geological disaster risk period of treatment.The analysis results show that the random forest model has the highest recognition accuracy of 94.89% in the classification of different local deformation pipe segments.With the increase of the length of intercepted sample data,the accuracy of 1D CNN model showed a trend of first increasing and then decreasing,reaching 93.84% when the length was 18 m.In the deep learning modeling,it is found that girth weld information played an obvious role in distinguishing different local deformation pipe segments,which could improve the overall identification accuracy of the model by 2%.The top 10 dangerous pipeline determined by the evaluation method has the coexistence of bending deformation,large displacement and metal loss.In summary,the method presented in this thesis can meet the engineering needs of IMU strain data analysis and evaluation.