Scholarlink

<p>The operating costs of urban water distribution networks (WDNs) are mainly due to pump stations. These costs involve energy consumption and pump maintenance, which can be reduced by optimizing pump operation. Pump scheduling can be reliable when the water demand for that period of pump operation is accurately predicted. This study focused on developing a pump scheduling system for a 24-h period that can be applied throughout the week based on water demand predictions. A deep four-channel convolutional neural network (4-channel CNN) model was introduced and implemented to forecast the hourly water demand for a weekly (168-h) horizon. This model achieved a mean absolute percentage error (MAPE) of 4.04%. The optimization was done using the Non-Dominated Sorting Genetic Algorithm, version II (NSGA-II), with the dual objectives of minimizing energy consumption and pump maintenance. The optimal Pareto Front was determined based on observed water demand values, demand values from the same hours the previous week, and predicted values obtained from one-dimensional CNN (1D CNN) and 4-channel CNN models. The results showed that using the 4-channel CNN model to forecast water demand and then incorporating this prediction into the optimizer model resulted in an MAPE of 3.29% compared to the observed demand on the optimal Pareto Front. In contrast, using the water demand values from the same hours last week and the demand predicted by the 1D CNN model resulted in an error rate of 23.97 and 16.25%, respectively. This highlights the importance of using a highly accurate forecasting model for pump scheduling optimization in WDNs.</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

Pump scheduling optimization in water distribution networks, including short-term demand forecasting by deep learning

  • 期刊论文

  • 发表日期：

    2025年12月5日

查看全文
  • Hossein Namdari,
  • Amin E. Bakhshipour,
  • Seyed Mohammad Ashrafi,
  • Ali Haghighi

摘要

The operating costs of urban water distribution networks (WDNs) are mainly due to pump stations. These costs involve energy consumption and pump maintenance, which can be reduced by optimizing pump operation. Pump scheduling can be reliable when the water demand for that period of pump operation is accurately predicted. This study focused on developing a pump scheduling system for a 24-h period that can be applied throughout the week based on water demand predictions. A deep four-channel convolutional neural network (4-channel CNN) model was introduced and implemented to forecast the hourly water demand for a weekly (168-h) horizon. This model achieved a mean absolute percentage error (MAPE) of 4.04%. The optimization was done using the Non-Dominated Sorting Genetic Algorithm, version II (NSGA-II), with the dual objectives of minimizing energy consumption and pump maintenance. The optimal Pareto Front was determined based on observed water demand values, demand values from the same hours the previous week, and predicted values obtained from one-dimensional CNN (1D CNN) and 4-channel CNN models. The results showed that using the 4-channel CNN model to forecast water demand and then incorporating this prediction into the optimizer model resulted in an MAPE of 3.29% compared to the observed demand on the optimal Pareto Front. In contrast, using the water demand values from the same hours last week and the demand predicted by the 1D CNN model resulted in an error rate of 23.97 and 16.25%, respectively. This highlights the importance of using a highly accurate forecasting model for pump scheduling optimization in WDNs.