Double-Layer Data-Driven Risk Assessment and Prediction of Drainage Pipeline Networks

Sewage pipeline network accidents occur frequently, posing a serious threat to the life and property safety of residents. There is an urgent need for methods that can effectively identify the key factors of pipeline risk as well as accurately assess that risk. Therefore, this paper proposes a two-layer data-driven risk prediction model for drainage pipe networks. The model utilizes a two-layer machine learning algorithm to overcome the challenges faced by previous pipeline risk prediction models, such as data deficiencies, and significantly improves the predictive performance of the model. In this study, a series of training, testing, and validation of four machine learning algorithms, namely, random forest, decision tree, logistic regression, and extreme gradient augmented tree, were carried out to compare the predictive performance of the model using a confusion matrix, accuracy, and checking completeness as well the area under the receiver operating characteristic curves. The results show that (1) the random forest model has the best prediction performance, and the extreme gradient boosted tree and decision tree models also have acceptable performance. The logistic regression model has relatively poor performance; (2) sewage discharge volume and surface vegetation cover are the key factors affecting pipeline risk; and (3) the number of high-risk pipeline segments in the study area is 602, accounting for 16.32% of the overall number, which is mainly concentrated in the vicinity of large-scale commercial and supermarket complexes. Medium-risk pipelines accounted for 14.64% of the overall, mainly distributed in densely populated areas. The model provided in this paper contributes to the prevention and control of drainage pipe risks and is instructive in modeling the characteristics of drainage pipe network risks using various machine learning algorithms.