A novel adaptive AQM scheme based on Neuron Reinforcement Learning (NRL) was presented. This scheme uses queue length and link rate as congestion notification to determine an appropriate drop/mark probability, and the parameters of neuron can be adjusted online according to the timevarying network environment, so that the stability of queue dynamics and robustness for fluctuation of TCP loads are guaranteed. This scheme is easy to implement with simple structure, and it is independent of the system model to be controlled. Simulation results show that the proposed algorithm can be effective in solving congestion control problem for complex network with uncertainties, and it has better performance on stability and robustness.