The opposing jet technique is a significant approach utilized in the thermal protection design of hypersonic vehicles. This paper aims at proposing the use of deep learning to control the flow rate of the opposing jet, to realize the intelligent and dynamic thermal protection process of the opposing jet. Using OpenFoam to study the thermal protection characteristics of an unsteady opposing jet on the surface of a blunt body, it is found that the recirculation zone formed by the flow field of the opposing jet plays a key role in the thermal protection leading to a certain delay of effect. Deep reinforcement learning is used to train the intelligent agent to control the opposing jet. By designing the reward function based on the thermal protection effect and the thermal protection efficiency, the intelligent agent is required to find a suitable thermal protection strategy. The thermal protection effect of the opposing jet controlled by an intelligent agent trained through deep reinforcement learning is 16% better than that of the sinusoidal opposing jet, and its thermal protection efficiency is 5.7% higher. It shows that the intelligent agent trained by deep reinforcement learning has good control effect on thermal protection of opposing jet.