Scholarlink

<p>This study explores the development and deployment of a predictive model for pipe failures within Barcelona's Water Distribution System. Using the XGBoost algorithm, refined through systematic analysis of explanatory variables, machine learning algorithms, and hyperparameter configurations, the model predicted up to 30.2% of expected pipe failures with a 5% annual renewal rate and 10.32% with a 1% renewal rate. Material-specific responses to predictive variations were observed, with ductile iron and HDPE pipes showing significantly different behaviours compared to non-cylinder reinforced concrete and fibre-cement pipes. The model development process was heavily guided by domain expertise, as reflected in the custom-built dataset, which was meticulously created considering local system conditions, strategic hyperparameter tuning, and optimisation using business-focused metrics. A preliminary exploration employing SHAP (Shapley Additive exPlanations) assessed the importance of local explanatory variables, which varied across pipe materials. This research advances understanding of how specific materials within water systems respond to predictive modelling of pipe failures, emphasising the vital role of extensive historical data in enhancing predictive accuracy and informing infrastructure planning and management.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Near-future prediction of pipe failures in water supply networks: a key determinant for water pipe renewal policies through a machine learning approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月8日

查看全文
  • Edwar Forero-Ortiz,
  • Marti Sanchez-Juny,
  • Eduardo Martinez-Gomariz,
  • Jaume Cardus Gonzalez,
  • Fernando Cucchietti,
  • Ferran Baque Viader

摘要

This study explores the development and deployment of a predictive model for pipe failures within Barcelona's Water Distribution System. Using the XGBoost algorithm, refined through systematic analysis of explanatory variables, machine learning algorithms, and hyperparameter configurations, the model predicted up to 30.2% of expected pipe failures with a 5% annual renewal rate and 10.32% with a 1% renewal rate. Material-specific responses to predictive variations were observed, with ductile iron and HDPE pipes showing significantly different behaviours compared to non-cylinder reinforced concrete and fibre-cement pipes. The model development process was heavily guided by domain expertise, as reflected in the custom-built dataset, which was meticulously created considering local system conditions, strategic hyperparameter tuning, and optimisation using business-focused metrics. A preliminary exploration employing SHAP (Shapley Additive exPlanations) assessed the importance of local explanatory variables, which varied across pipe materials. This research advances understanding of how specific materials within water systems respond to predictive modelling of pipe failures, emphasising the vital role of extensive historical data in enhancing predictive accuracy and informing infrastructure planning and management.