The escalating frequency of extreme weather events and the emergence of recalcitrant pollutants have rendered traditional water quality management paradigms—predicated on periodic sampling and steady-state mechanistic models—increasingly inadequate, often yielding prediction errors exceeding 40% under non-stationary hydrological regimes. To address this, we reconceptualize water quality not as discrete physicochemical attributes, but as a time-domain signal function, , thereby establishing a mathematical bridge between physicochemical processes and information-theoretic control theory. Grounded in this signal-based redefinition, we propose a novel five-layered resilient architecture: comprising Stereoscopic Perception, Data Fusion, Intelligent Modeling, Collaborative Decision-making, and Interactive Application layers, specifically designed to operationalize the "Water Quantity-Quality-Ecology" (3W) synergistic governance. Unlike conventional digital twins that merely simulate evolution, our framework emphasizes a closed-loop feedback mechanism that integrates multi-objective optimization(minimizing operational costs and carbon footprints while maximizing pollutant removal) with adaptive control strategies. Deployed in a full-scale wastewater treatment plant expansion, the resilience-oriented regulation achieved approximately 20% reduction in aeration energy consumption and a significant decrease in indirect carbon emissions, even under influent load shocks. Furthermore, we establish a multi-dimensional resilience evaluation system via entropy-weight logistic regression to quantify systemic robustness. This work provides a theoretical underpinning and actionable pathway for transitioning water pollution control from passive "end-of-pipe" compliance toward proactive, intelligence-driven ecological sustainability.