The invention provides an automatic parking control system. The whole automatic parking process is divided into an input layer, a policy layer, a planning layer and a control layer by the system. In an automatic parking control process, the input layer generates control instruction and parking track data sets of vehicles in different types by virtue of simulation; the policy layer learns the simulation data by utilizing a deep neural network algorithm so as to extract the common relationship between control instructions and parking tracks; the planning layer finds a proper parking policy by virtue of training of few steps to give a control instruction for the parking scene, thereby generating a planning track; and the control layer performs control feedback according to the deviation of anactual parking track and the planning track so as to enable the parking track to be most similar to the ideal track planned by the system.