Artificial intelligence for decentralized water systems: a smart planning agent based on reinforcement learning for off-grid camp water infrastructures

The planning and management of decentralized technologies in water systems is one of the promising, yet overlooked, domains where artificial intelligence (AI) can be successfully applied. In this study, we develop and deploy a reinforcement learning (RL)-based 'smart planning agent' capable of designing alternative decentralized water systems under demanding operational contexts. The agent's aim is to identify optimal water infrastructure configurations (i.e., proposed decisions on water management options and interventions) for different conditions with regard to climate, occupancy and water technology availability in a demanding, off-grid setting, i.e., a water system with high requirements of independence from centralized infrastructure. The agent is coupled with a source-to-tap water cycle simulation model (Urban Water Optioneering Tool, UWOT) capable of assessing and stress-testing the proposed configurations under different conditions. The approach is demonstrated in the case of a military camp deployed abroad for peacekeeping operations. The agent is tasked with selecting optimal interventions from an array of real-world camp water management technologies and evaluating their efficiency under highly variable, operational conditions explored through simulation. The results show that RL can be a useful addition to the arsenal of decision support systems (DSS) for distributed water system planning and management, especially under challenging, highly variable conditions.

Automated summary

In this study, a reinforcement learning-based smart planning agent is developed to design decentralized water systems under challenging operational contexts. The agent is coupled with a simulation model to assess and test proposed configurations, and it demonstrates good performance in a military camp case study.