Study on Gamma selection in the optimal operation of secondary water supply system based on deep Q-learning network

When the requirements for water pressure and quantity of drinking water for residents and industrial buildings exceed the capacity of the urban water distribution system, a secondary water supply system (SWSS) is supplied to users by pipelines through storage, pressurization, and other facilities. In China, SWSS has been installed in 97% of residential buildings and the operation of SWSS is directly related to the water pressure and water quality of the users' tap water. In this paper, the operation optimization objectives for the SWSS with storage facilities were proposed, and deep Q-learning network (DQN) was applied to optimize the control of SWSS. In this study, the pressure, the water age in the roof water tank, and the power consumption of the pumps were selected as the optimization objectives. The equation for the qualitative selection of the key hyperparameter (Gamma) was proposed and verified by the experiments in a community of City S in East China. The results indicated that with the decrease in the volume of the water tank, the larger Gamma value was recommended, and the more future conditions were considered. It is hoped that the result can be used as a reference in SWSS operation optimization.