This paper proposes three different network traf? c prediction models based on deep belief network to predict the Internet traffic in the next 1 hour. First

the network structure of deep belief network is introduced. Then

the deep belief network topology of three different architectures was constructed. Finally

through experimental comparison

it was found that the number of neurons in the hidden layer was crucial to the deeper level of the network. This model proved to be an effective prediction model. The method adopted in this paper provides accurate network traf? c prediction while simulating traf? c data patterns and random elements

so that the root mean square error value of the test data set is 0.028.