Constrained Dynamic Matrix Control of HRSG Water–Steam Chemistry Under Measurable Disturbances

Heat recovery steam generator (HRSG) water–steam chemical dosing involves transport delays, multivariable coupling, measurable disturbances, measurement uncertainty, and operating constraints. This study develops an engineering-inspired constrained dynamic matrix control (DMC) benchmark for HRSG feedwater and condensate ammonia dosing using a three-controlled-variable (CV), two-manipulated-variable (MV), and five-measurable-disturbance-variable (DV) first-order-plus-dead-time (FOPDT) model. The controller incorporates measurement-based prediction correction, current measurable-disturbance compensation, quadratic programming (QP), receding-horizon optimization, and predefined proportional–integral (PI) fallback logic. Performance is evaluated using 30 matched Monte Carlo scenarios against independently tuned diagonal- and cross-paired PI baselines, with additional centralized-PI, disturbance-channel, measurement-quality, constraint-feasibility, and soft-CV diagnostic audits. DMC with zero-order-hold disturbance prediction reduces high-pressure (HP) and intermediate-pressure (IP) pH integral absolute error by 9.61%/13.61% relative to PI-diagonal and by 18.32%/10.17% relative to PI-cross. Explicit use of current measurable disturbances further reduces HP/IP pH error by 3.11%/2.02% compared with DMC without disturbance-variable information, whereas short-window trend extrapolation provides no statistically supported additional tracking benefit after Holm adjustment and increases MV1/MV2 total variation by 16.59%/86.06%. Under an extreme conductivity-pressure condition, the original hard-constrained DMC becomes finite-horizon infeasible for most attempted QPs even though separate steady-state feasibility checks are satisfied; a diagnostic soft-CV formulation largely removes Phase-I infeasibility by accepting penalized prediction-domain slack. The FOPDT dynamics are engineering-inspired benchmark assumptions rather than parameters identified from plant historian or step-test data.