With the rapid development of the tourism industry, traditional travel route planning methods rely on static rules or simple optimization algorithms, which are difficult to handle user preferences and dynamic environmental changes (such as traffic congestion and sudden weather changes) in real time, resulting in insufficient route personalization and low efficiency. To this end, this study proposes an intelligent planning algorithm based on deep reinforcement learning, which aims to optimize user time cost and experience satisfaction through dynamic decision-making. The algorithm first models the travel route as a Markov decision process, defines the state space (user location, historical preferences, scenic spot opening hours, real-time traffic), action space (attraction selection) and reward function (integration of satisfaction score, travel time and route coherence). Secondly, a deep Q network (DQN) is constructed, using a dual network architecture (online network generates Q value, target network calculates target value) and an experience replay mechanism to solve the training oscillation problem caused by data correlation. Finally, a multi-head self-attention mechanism is introduced to dynamically capture the association weights between user preferences and scenic spot features. In the comparative experiment of Ctrip data set and simulation environment, the trip time of this algorithm is reduced to 3.9 h at the lowest, and the response adjustment time in the traffic mutation scenario is as low as 1.6 s. The results show that the algorithm effectively balances personalized needs and dynamic constraints through the deep reinforcement learning framework, providing a highly adaptable solution for smart tourism services.