A bilevel data-driven method for sewer deposit prediction  under uncertainty

Wenli Liu , Yexin He, Zihan Liu , Hanbin Luo, Tianxiang Liu

ABSTRACT

Deposit accumulation is one of the predominant causes of sewer blockage and overflow. Nevertheless, the traditional detection methods are costly and time-consuming, and the accuracy of the mathematical models for deposit prediction is usually affected by some uncertain factors (e.g., pipe properties and flow velocity of water). This paper proposes a framework of global sensitivity analysis (GSA) to identify the most sensitive indicators for sewer deposit prediction by (i) developing a data-driven bilevel (i.e., catchment level and segment level) model to map the relation between input and output indicators and (ii) employing three different GSA methods, namely, the Morris method, Sobol method, and Borgonovo index method to identify the indicators as important or unimportant (insensitive). The results show that the likelihood of combined sewer overflow occurrences (LCSOO), pipe age (PA), and pipe material (PM) are influential parameters for the thickness of deposits. Here, we pay close attention to the most influential parameters, which can help improve forecast prediction accuracy.

Keywords: Sewer system,Generalized linear mixed modeling (GLMM),Polynomial-Chaos Kriging (PC-Kriging),Sewer deposits,Global sensitivity analysis (GSA)

https://www.sciencedirect.com/science/article/pii/S0043135423000246