Planning Flexible Water Supply Networks Under Uncertainty Using Reinforcement Learning: Insights from a Real-World Case Study

<p dir="ltr">Drinking water utilities are under increasing pressure to provide sufficient water to their customers in a sustainable and cost-effective manner, amid a rapidly changing and uncertain environment. Climate change, ageing infrastructure, population growth, and evolving regulations are rendering long-term planning of water distribution systems increasingly complex. These challenges necessitate more adaptive planning approaches that can respond to emerging information and develop flexible strategies capable of performing effectively across a wide range of possible future scenarios. This need is particularly critical in the context of strategic supply networks, which connect water sources to demand centres and represent a significant investment for utilities. This paper explores the practical application of Deep Reinforcement Learning (DRL) as a decision-making framework for strategic water supply network planning. We apply our approach to a real-world case study, optimising network configuration and resource allocation. By benchmarking our approach against expert-designed solutions, we demonstrate its real-world effectiveness and highlight the tangible benefits of incorporating optimal decision-making techniques. This contributes to the development of a new generation of dynamic planning decision support tools for the water sector.</p><p dir="ltr">This paper was presented at the 21st Computing and Control in the Water Industry Conference (CCWI 2025) at the University of Sheffield (1st - 3rd September 2025).</p>