SMART STOCHASTIC CONTROL OF CLOSED WATER SYSTEMS VIA RELAXATION-BASED HYDRODYNAMIC MODELS

Abstract

This study tackles the challenge of automated control in closed water systems—such as endorheic lakes—that experience stochastic hydrological fluctuations and internal relaxation effects (non-Markovian dynamics). We introduce an intelligent stochastic control approach using Relaxation-Stochastic Models (RSMs), which account for both probabilistic environmental inputs and system memory. A risk-aware optimal control problem is solved online through Receding Horizon Control combined with an LSTM-based metamodel of the RSM. When applied to Uzbekistan’s Aydar Lake system, the proposed method reduces the likelihood of violating critical ecological thresholds by 37% over a 10-year period, while boosting irrigation supply reliability by 18% relative to conventional methods. These findings offer a robust and adaptive pathway for sustainable water resource automation under deep uncertainty