Risk Assessment Of Municipal Pipe Network Operation And Maintenance Based On Machine Learning

As an important part of underground pipeline,the scale of municipal water supply and drainage pipeline is expanding.Its intelligent management and operation and maintenance level are directly related to the safe operation of the city’s comprehensive pipe network.The traditional risk assessment of pipeline network is mostly the semi quantitative analysis of accident consequence,which depends on the accuracy of hydraulic model and the reliability of numerical simulation.It can not match the development of the quality and quantity of available data in the pipeline system which is brought by the improvement of multiple monitoring means.Therefore,this paper proposes a risk assessment method of municipal pipe network operation and maintenance based on machine learning,which is driven by data,with lower cost and higher labor and time efficiency.Once calibrated,optimized and verified,there is no need for too many operators,mainly by the computer system work,risk assessment algorithm for continuous monitoring of the pipeline system,making the maintenance of the pipeline network model become very efficient.First of all,collect the research data of operation and maintenance risk of municipal pipe network at home and abroad.Through data mining,literature review,inquiry research methods,on the basis of statistics,this paper studies the content of urban municipal pipe network operation and maintenance risk and the causes of accidents,including three types of physical,operational and environmental characteristics.Sorting out the safety accident factors in the operation and maintenance process of municipal pipe network,it is found that the proportion of corrosion and aging in the accident types of water supply and drainage pipe network is relatively large,followed by the construction quality or third-party damage factors.Secondly,in view of the shortcomings of the existing pipeline network operation and maintenance risk assessment model,a machine learning based pipeline network operation and maintenance risk assessment method is proposed,including the mathematical principle of the model,data preprocessing method,parameter selection and optimization.By evaluating the learning performance of several algorithms,the logical regression and random forest algorithm are established.Thirdly,an example of operation and maintenance risk assessment model of municipal pipe network is established.Based on the data of a park network base in Suzhou,the model is established.Determine the source and type of monitoring data,preprocess and balance the data,ensure the data quality and sampling accuracy;build two learning models based on single learner logical regression method and integrated learner random forest method based on the data structure and rules preliminarily determined by exploratory data analysis,and evaluate the learning effect of the two methods.The risk probability related to each pipe section sample is obtained,and the risk factors affecting the failure of pipe section are determined,and their correlation and importance order are determined.The results show that the most influential factors are pipe material,soil property,service life and previous failure times.The random forest algorithm has better prediction accuracy and robustness in this data set.Finally,an open-source,platform independent,closely related GIS and distributed network risk assessment modeling framework is proposed to improve model data integration.By developing an integrated user interface that shares the geographic data model with the underlying layer,the data flow between the platform processing components is improved.The framework and method can be applied to different risk prediction models,and provide reference for future GIS model integration.