In order to tackle traffic congestion and safety issues in expressway merging areas and to ensure efficient, safe, comfortable, and stable travel of connected and automated vehicles (CAVs) in these areas, this study employs the DQN (deep q-network) algorithm from deep reinforcement learning. By considering factors such as vehicle safety, efficiency, and comfort, a reward function model for neural network training has been established, and a CAV lane-change decision-making method for merging areas has been proposed. Using the open-source highway-env simulation scenario, a simulation environment for expressway merging areas has been set up, and experiments have been conducted on the mainline and ramps. The results of the simulation experiments show that compared to the intelligent driver model (IDM) and the lane-change decision-making method in highway-env, the proposed CAV lane-change decision-making method enables CAVs to quickly reach a stable driving state at a speed of 22.22 m/s. It also reduces frequent lane changes and acceleration/deceleration behaviors, and optimizes the time-headway between vehicles. This significantly improves the efficiency of traffic flow and ride comfort. The research findings provide a new method for vehicle traffic management in urban expressway merging areas under intelligent networked conditions. They also offer a decision-making approach for lane changes in future connected and automated vehicles.