The invention discloses a short-time parking demand prediction method based on a GRU model. The short-time parking demand prediction method comprises the following steps: 1), obtaining historical dataof parking lot facilities, processing the historical data, and obtaining the parking space occupancy data at each time point; 2) setting a GRU neural network structure by using a deep learning Kerasframework packet, and obtaining model optimal parameters by using a GridSearch function in the Keras packet; and 3) training a GRU model by using the training set data, storing the model and predicting the berth occupancy of the next step length. Compared with the prior art, under the background of obtaining continuous parking data, the short-time parking demand prediction method utilizes the bigdata processing technology, applies the latest algorithm of deep learning, provides a more advanced and more accurate parking information induction publishing method, and can consider the relevance ofthe parking demand in the time dimension while the cell module has a simpler control door structure, so that training efficiency can be improved to a great extent, and then the utilization rate of parking facilities is increased, and satisfaction of users with parking requirements is improved, and unnecessary traffic of the users is avoided, and traffic pressure of roads is reduced, and effectivetraffic control of traffic management departments during peak hours of traffic flow is facilitated.