Abstract: A method for a self-adaptive service function chain mapping based on a deep reinforcement learning, comprising: establishing an SFC mapping model, dividing an SFC mapping process into a three-layer structure, and representing the structure with abstract parameters; building an SFCR mapping learning neural network, and mapping the abstract parameters to a state, an action and a reward value in the SFCR mapping learning neural network; establishing an empirical playback pool and updating network parameters; summarizing request rates and utilization rates of different VNFs, a number of currently deployed VNFs and a number of unactivated VNFs based on data in the empirical playback pool; and designing a VNF redeployment strategy, and redeploying the VNFs according to the summarized data. The method has good self-adaptability, and can improve the effective service cost rate and the request mapping rate for processing service requests from users in different time periods.