Traffic matrix estimation is an interesting research problem at present.Network operators use it to conduct load balancing

route optimization

traffic detecting

network dimensioning and so on.However

the highly ill-pose nature of traffic matrix estimation itself makes it being a challenging research subject to estimate accurately traffic matrix.This paper studies traffic matrix estimation in large-scale IP backbone networks.Based on RBF (radial basis function) neural network

a novel estimation method

namely TMRI (traffic matrix recurrence inference)

is proposed.TMRI exploits the powerful modeling ability of RBF neural network to model traffic matrix estimation.The ill-pose nature of this problem will be overcome in the process of training the RBF neural network.Accordingly

the complex process of mathematic modeling can be avoided.Built on this estimation model

traffic matrix estimation is described into the optimal process under the constraints.By seeking the recurrent optimal solution

TMRI can further get rid of the ill-pose nature of this problem.Simulation results show that TMRI can accurately estimate traffic matrix and track its dynamics

and in contrast to previous methods

it holds the stronger robustness to noise and more evident performance improvement.