Equitable supply in water distribution networks using data-driven optimization and AI-assisted voice automation

Intermittent water supply is common across the Global South. In Multi-Village Schemes, one source supplies several village storage tanks within fixed supply periods, making balanced allocation difficult when flow information is limited. This study presents an integrated framework for multilingual demand input, data-driven scheduling, and automatic valve operation. A voice interface collects tank demands, the scheduling model generates a valve schedule, and a programmable logic controller executes the schedule without manual transfer between stages. The framework adapts an existing discrete-event model, mixed-integer linear programming formulation, and ε -greedy state-selection strategy. It uses measured state-specific flows instead of hydraulic equations. On an eight-tank laboratory network with 256 valve states, flow measurements from 33 states (13%) produced a schedule with an absolute solver-predicted deviation below 2.6% for every tank. Across eight physical trials, the largest absolute mean experimental deviation was 5.55%. The optimized schedules produced more balanced allocations than the basic operating policies. In simulation, ε -greedy and pure greedy performed similarly for the eight-tank network, while ε -greedy required fewer states on average for the ten-tank network. The voice interface achieved 97–98% entity-extraction accuracy under the tested clean and noisy conditions, with no hallucinated demand assignments observed. These results demonstrate an integrated demand-to-actuation workflow on a laboratory-scale intermittent water-supply testbed.