In intent-driven optical networks, due to the sudden increase in the number of intents sent at the same time or the limitation of underlying physical resources, some users' intent requests cannot be successfully connected, resulting in intent conflicts. In this paper, an intent-driven optical network conflict resolution method based on reinforcement learning is proposed. According to the user intention rollback policy output by reinforcement learning, the network demand index corresponding to the user intention is updated, and the average quality of service（QoS） value in the current network is maximized on the premise of ensuring that the user intention can be successfully connected. The experimental results show that the proposed method can effectively improve the average QoS value of users in the current network environment and reduce the network blocking rate.