Cognitive networks can perceive the external environment

and intelligently and automatically change their behavior to adapt to the environment

so it is more suitable to provide users security with QoS.Designing a high-precision traffic prediction model can improve the cognitive feature of cognitive networks.For the models of low forecasting precision

highly independence to training samples and bad description of network traffic

we propose a new model.It trains BP with Ant Colony Algorithm

which avoids slow convergence speed and easily falling into local optimum problems existed in the gradient descent method.Besides

we reject the abnormal data using BP1

do wavelet decomposition

and predict the network traffic with hybrid model.Experimental results show its high-precision in cognitive networks.