The invention discloses an edge cooperation caching method under a hierarchical wireless network supporting D2D communication. The method comprises the steps that whether a local cache of a device receiving a content request contains request content or not is checked according to the content request; if yes, the request content is hit; if not, routing selection is performed based on a deep reinforcement learning algorithm to hit the request content; and after the request content is hit, the model of the deep reinforcement learning algorithm is updated. According to the method, a deep reinforcement learning algorithm is applied to collaborative cache configuration of a user equipment layer in a mobile wireless network; the unsupervised learning of the cooperative cache of the user equipment is realized, the hit rate of the maximum cache is taken as an optimization target, and the flow is unloaded to the periphery of the user as much as possible, so that the flow transmission in the cellular network is greatly reduced, and the transmission delay of the user request content is shortened.