Evaluation of data driven models for pipe burst prediction in urban water distribution systems

ABSTRACT

The water distribution network is one of the most expensive parts of a water supply system. The fundamental variables of a network, material, diameter, length, age, and the hydraulic pressure of pipes are the factors that affect the pipe burst rate (PBR). Establishing a relationship among the burst rate and these factors is an important step to assess the conditions governing the network and preventing significant water leakage. Implementing the data-driven approach in PBR prediction is an effective method to find the relationship. In the present study, Grasshopper Optimization Algorithm-based Support Vector Regression (GOA-SVR), Gaussian Process Regression (GPR), and Artificial Neural Network (ANN) have been developed to predict PBR in an urban area. The results show that the GPR model outperforms other methods. Furthermore, the sensitivity analysis indicates that the pipe age has a negative effect on PBR modeling while the pipe length is the most relevant variable.

KEYWORDS:

• Data-driven models
• machine learning
• pipe burst rate
• water distribution systems

Disclosure statement

No potential conflict of interest was reported by the authors.