The invention discloses a method of controlling a flow bandwidth. For one data flow set, under the condition that the data flow set uses a historical adjusting strategy and a predefined strategy to distribute labels in a historic time period, a flow state satisfies an expected degree and a machine learning model is trained, and the machine learning model acquired through training determines a current adjusting strategy which makes a subsequent flow state of the data flow satisfies an expectation maximumly according to a flow statistics result of the data flow set in a current time period. Therefore, network equipment can adjust a strategy which is predefined by a user according to the changed flow state so that labels distributed for a key service are increased and/or labels distributed for a non-key service are decreased. A sending speed of a non-key service flow is reduced so that a link bandwidth is not occupied excessively by the non-key service flow and a data flow of the key service can occupy an enough bandwidth at an entrance and an exit of network equipment.