The invention discloses a central air-conditioning control method based on multi-agent deep reinforcement learning, and the method comprises the steps: carrying out the model-free optimization control of the starting and stopping states and working parameters of a cooler, a cooling water pump and a cooling water tower fan in a central air-conditioning system according to the current indoor demand cooling load and outdoor wet bulb temperature, including the operation sequence control of the cooler; according to the control method, an accurate central air-conditioning system model does not need to be established in the actual deployment process, the working frequency of the cooling water pump and the working frequency of the cooling water tower fan can be respectively controlled only by using a single agent, and the working frequency of the cooling water pump and the working frequency of the cooling water tower fan can be controlled by relying on a small amount of historical data. An efficient and accurate control strategy is trained in a short time, the unnecessary refrigerating capacity is reduced, the workload of a refrigerator, a cooling water pump and a cooling water tower fan is reduced, the service life is prolonged, the failure rate is reduced, and the energy consumption of the whole central air-conditioning system and even the total energy consumption of a building are greatly reduced.