Artificial neural network-based pump operation policy models for energy cost minimisation

Files

hdl_150066.pdf (13.46 MB)
  (Published version)

Date

2026

Authors

Zheng, L.
Wu, W.
Simpson, A.R.
Wang, Y.

Editors

Advisors

Journal Title

Journal ISSN

Volume Title

Type:

Journal article

Citation

Engineering Applications of Artificial Intelligence, 2026; 167:113874-1-113874-14

Statement of Responsibility

Lang Zheng, Wenyan Wu, Angus R. Simpson, Ye Wang

Conference Name

Abstract

Efficient pump operation in water distribution systems (WDSs) is crucial for meeting end-users’ demands while minimising energy costs. Artificial neural network (ANN)-based operation policy models have been developed to address the limitations of existing methods in adjusting pump operations under dynamic conditions, where water demand and electricity tariffs vary. However, developing a suitable ANN-based policy model for a specific WDS is challenging due to the distinctly different learning patterns of ANN architectures and different operational characteristics of fixed and variable speed pumps, which collectively influence model performance. In this study, a modelling framework is proposed that couples a physics-based hydraulic model with an ANN-based operation policy model to inform decision-making under dynamic conditions. Two different ANNs, namely the radial basis function (RBF) and the multilayer perceptron (MLP), have been developed as policy models, and their strengths and limitations were evaluated through two WDSs - one with fixed speed pumps and another with a variable speed pump. The ANN-based policy models have been evaluated for their ability to minimise pumping energy costs and detailed operational behaviour. For fixed speed pumping system, the RBF-based policy model outperforms the MLP-based policy model in minimising energy costs and exhibits more advantageous operational behaviour under conditions driven by peak daily demand. For variable speed pumping system, both ANN-based policy models perform reasonably well in minimising energy costs. The RBF-based policy model remains more advantageous under conditions with peak daily demand, while the MLP-based policy model can be more advantageous under conditions with high electricity tariffs.

School/Discipline

Dissertation Note

Provenance

Description

Access Status

Rights

© 2026 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

License

Call number

Persistent link to this record