The invention discloses an air control method based on reinforcement learning and a four-dimensional trajectory. The method comprises the following steps: firstly, establishing aerodynamic performance models of aircrafts of different types; acquiring four-dimensional trajectory data of different airlines for different airlines according to the aerodynamic performance model of the aircraft; through data playback, generating a four-dimensional trajectory model of an airline-aircraft model; and finally, based on a reinforcement learning algorithm, establishing a neural network, training a four-dimensional trajectory on the movement of the aircraft, constructing a nested reinforcement learning model of a nested speed agent in a course agent, and selecting an aircraft route by choosing a target course of the aircraft. The arrival time of the aircraft is controlled by selecting the target speed of the aircraft, and therefore the function that the aircraft presses the four-dimensional trajectory model according to the specified time, speed, course and height is achieved. According to the invention, a feasible solution can be provided for the problems of large flow, complex aircraft scheduling method, difficulty in air control and the like of the current airport.