An optimal operation model of an urban drainage system under partial state observation

Major issues of urban drainage are to cope with flooding caused by torrential downpour and water pollution caused by washing load from non-point sources. To reduce the damage of flooding, large-scale trunk sewer and pumping stations are constructed in large Japanese cities. To reduce the damage due to water pollution, these sewerage systems should have additional functions of storage and treatment of the washing load. In such a case, pump operation has key roles for reduction of damages. To obtain the optimal policy of such operation, many types of optimization method have been proposed. Stochastic dynamic programming (SDP) is an excellent method of obtaining the optimal operation policy. Most SDP models assume the state variables such as storage volumes are perfectly observable. However, for actual operation, construction of the observation system to observe state variables perfectly is difficult. Therefore, these facilities should be controlled under imperfect information. The research in this field is not so advanced though such problems are thought to be important in a city, which has many such facilities. The paper shows formulation of such optimal control under imperfect (partially observed) information by using a SDP model.