As an important waterway in China, the Three Gorges Reservoir Area faces complex challenges in water level control during the change period (such as the transition stage between flood season and dry season or the adjustment period of water conservancy project scheduling), including uncertainties in incoming water, conflicts in multi-objective scheduling, and dynamic changes in the hydrological environment. Traditional water level control methods are difficult to achieve dynamic adaptive adjustment, leading to potential risks to the navigation efficiency and safety of the waterway. This paper proposes an optimization method based on Deep Double Recurrent Q-Network (DDRQN) for the adaptive control of water levels in the Three Gorges Reservoir Area waterway during the change period. This method combines the decision-making ability of deep reinforcement learning with the time-series information processing ability of recurrent neural networks, and learns the optimal control strategy through continuous interaction with the environment. Experimental results show that compared with traditional PID control and DQN methods, the DDRQN method has significant improvements in water level control accuracy, dynamic response speed, and anti-interference ability. It can effectively adapt to the complex changes in the hydrological environment of the reservoir area during the change period, providing new technical support for ensuring the safety and efficient navigation of the waterway.