Leveraging Drinking Water Pumps as Flexible Loads Using Input Convex Neural Networks
Document Type
Conference Proceeding
Publication Date
1-1-2025
Abstract
Drinking water distribution networks can be operated as flexible loads within the electric power grid due to their substantial pumping demands and water storage capabilities. Optimizing the flexible operation of a water distribution network poses significant challenges due to the complex physical laws within the network, where the hydraulic head difference equations for pipes and pumps are nonconvex. Standard nonconvex optimization solvers often fail to provide globally optimal solutions and the time required for computation can be prohibitively large. To resolve these issues, we present an optimization approach that accurately approximates the nonconvex constraints using input convex neural networks (ICNNs). This method converts the mixed-integer nonconvex optimization problem into a mixed-integer linear program, improving computational efficiency and scalability while maintaining the optimization problem's intuitive structure. In two case studies, we compare the ICNN-aided approach with the original nonconvex problem and found that the ICNN-aided approach outperforms the nonconvex solver in terms of computational time and optimality.
Publication Title
IEEE Power and Energy Society General Meeting
ISBN
[9798331509958]
Recommended Citation
Sakib, A.,
&
Stuhlmacher, A.
(2025).
Leveraging Drinking Water Pumps as Flexible Loads Using Input Convex Neural Networks.
IEEE Power and Energy Society General Meeting.
http://doi.org/10.1109/PESGM52009.2025.11225422
Retrieved from: https://digitalcommons.mtu.edu/michigantech-p2/2240