The present disclosure describes acquiring characteristic behavior data of a user in real time, and at a detection triggering moment, matching the characteristic behavior data with a behavior pattern baseline obtained through offline calculation, thus predicting a peak or a trough of resource use. In this way, a resource specification to be used by the user is dynamically adjusted in advance according to a behavior pattern of the user, to meet a resource use demand of the user, and avoid that the user is in shortage of resources in the peak time or wastes resources in the trough time, thus achieving better predictability.