Due to the fast development of deep reinforcement learning technology in recent years, the excellent decision-making ability provided by it has shown great success in various optimization problems. In this context, an intelligent joint optimization and control method based on the multi-agent technology is proposed for large-scale water conservancy hub gate-dam complexes in this paper. First, studies about intelligent gate control for dam gate complexes using reinforcement learning technology are reviewed. Using reinforcement learning algorithm, a multi-agent reinforcement learning model was built using real-time monitoring information and optimal gate control simulation information. Further using this model, this paper also proposed a decomposed multi-agent method to address multi-objective flood control scheduling problems for dam complexes. By decomposing the problem into several sub-problems, different reinforcement learning agents will be used to generate an optimal solution set. This approach was validated by using simulation including the process of actual flood changes when scheduling dams, calculating reservoir water storage amounts and comparing it with measurements. Through the coordinated regulation, the total storage capacity can increase by 7.19 million m³, 9.88 million m³, and 11.36 million m³ at Liuqiao, Gongjia, and Baoji sluice gates respectively. As the amount of floods increases, the storage capacity also grows proportionally, almost twice as much as the real-world storage amount.
