A Utility-Aware Probabilistic Digital Twin Framework With CC-MPC for Hard-Compliance Real-Time Decision Support in Continuous-Flow Environmental Treatment

Compliance-critical environmental treatment units must satisfy hard discharge limits while operating under time-varying disturbances and model uncertainty. This paper presents a utility-aware probabilistic digital twin (PDT) framework that couples online Bayesian uncertainty maintenance with chance-constrained model predictive control (CC-MPC) for minute-level decision support. The PDT maintains a posterior over uncertain process parameters using sequential Bayesian filtering, which induces calibrated predictive distributions for compliance-relevant variables. CC-MPC then selects dosing actions by enforcing chance constraints on the predicted discharge quality while minimizing expected operating cost. To enable reproducible and decision-relevant comparisons under practical data-confidentiality constraints, we introduce a compliance-gated utility-aware evaluation protocol: implementations must first pass explicit compliance gates before utility (cost) is compared, and calibration is reported as a necessary diagnostic. A numerical study based on anonymized/simulated minute-level signals, including an injected drift event, demonstrates the end-to-end workflow, reports compliance and cost outcomes, and quantifies computational load for real-time feasibility.