The Research Of Assessment And Prediction Methods For Pipeline Defects Based On Big Data Analysis

Pipeline is an important way for oil and natural gas transportation.In recent years,the operating mileage of China's pipelines has been increasing,meaning that pipeline safety risks also have been increasing.Serious accidents often occur with serious consequences.However,due to most of the pipelines are underground or undersea,it is unable to do the real-time monitoring.Therefore,the assessment and prediction of pipeline defects are much important in safety management.With the construction of digital pipeline,a multi-source mass with distributed storage mode is formed in pipeline's entire life cycle,with the characteristics of multi-category and large amount,which is known as pipeline big data.Thus,analysis based on big data methods has become an important development trend of the pipeline industry.Aiming at pipeline safety decision support,this thesis studies pipeline defects assessment and prediction mehods based on big data analysis,to solve the problem that may come during construction of pipeline big data and the research of big data analysis in assessment and prediction methods of pipeline defects in the future,and provide reference for the pipelines industry.Carry out pipeline defect prediction research based on multi-source data.This thesis puts forward solutions to the problem of information association,data alignment in the process of pipeline big data,and explores application of big data technology in pipeline defect prediction.First of all,against the problem that the presence of Chinese in the inline inspection data limits the implementation of automatic alignment,a semantic similarity model based on the Synonyms Word Forest is established.In this model,the international universal Miller & Charles data set is used to set the weights to ensure the universality of the model.The accuracy of the model is verified with the words in the actual in-line inspection report.Then,the depth prediction model of corrosion defects in pipelines using data-driven algorithms is studied based on the research results of in-line inspection alignment.This model is different from the existing mechanism-based prediction models for small samples,since it is not necessary to obtain real-time monitoring defect data and corresponding mechanism research,which is more suitable for field applications.At the same time,the Generalized Additive Model algorithm used in the model can also establish the relationship between the independent variables and the increase of corrosion depth,helping to analyze the influence of each factor.Carry out assessment study of pipeline defects.This thesis establishes a pipeline defects correlation model,and explores the application of big data technology in the optimization of traditional pipeline defect assessment methods.First of all,aiming at the non-linear characteristic of pipeline big data,this thesis studies the correlation model of pipeline defects based on Pearson correlation coefficient improved by information theory,which is used to mine the key factors.This model analyzes the correlation between conditional factors and decision factors from the perspective of information transportation to obtain more realistic results,which is different from the initial methods mainly based on linear relationships.Then,based on the correlation analysis model,taking the evaluation parameter in the international standard ASME B31 G,named Safety Factor,as an example,the application of big data analytical methods in the assessment of pipeline defects is studied.The Multi-Criteria Decision-Making method is used to modify the original calculation results,reducing the selection range of the Safety Factor to help decision-making.The impact of environment factors and defect information is reflected on the difference in Safety Factors,making it more realistic.