Many cities in the U.S. are faced with the problem of combined sewer overflows (CSOs). Combined sewers carry a mixture of municipal sewage and storm runoff. During storm conditions, the capacity of such sewers (and/or the associated treatment plants) may be exceeded and the combined sewage must be released to the environment untreated. Real-time control of the combined sewage conveyance system is a potential low cost solution to the problem. Regulating gates and pumps can be manipulated to maximize storage of peak loads within the system to minimize the CSOs during each storm event. The problem described above is the design of a control algorithm for a large-scale system. The PI has adapted Model Predictive Control (MPC) strategies and developed a control scheme for the Metropolitan West Seattle sewage collection system. The project has involved: (1) optimization of the dynamic response of a plant in which there are many manipulated variables and many output variables, (2) forecasting of unmeasured plant disturbances, (3) accounting for inequality constraints on most output variables, and (4) incorporating plant dynamics that include integrator states, nonlinear characteristics, and lags and time delays. This research is a one-year case study of the performance of MPC on the operation of the Seattle metropolitan sewage system. The application-specific goal is to minimize discharge of raw sewage from the network during storm events. Simulation studies predict that CSOs can be reduced by 28% under MPC (relative to the existing rule-based computer- control strategy). MPC software has already been installed, and will regulate flowrates at 23 locations in the network on a 10-minute cycle. Data will be collected automatically during each storm event in the 1992-93 rainy season, then analyzed off-line to: (1) measure the reduction of CSOs, and (2) diagnose any shortcomings in the MPC strategy using multivariate statistics. Data from the use of MPC on this large-scale problem should encourage application of MPC to similar problems which exist throughout the chemical industry. The results will also provide new insights on generic issues in MPC, including: (1) impact of (and compensation for) model error, unmeasured disturbances, and sensor/actuator failure, and (2) formulation of constraints and objective function-ease with which competing control objectives can be achieved in the MPC framework.