Active queue management algorithm based on data-driven predictive control

Model predictive control (MPC) is a popular strategy for active queue management (AQM) that is able to incorporate physical and user defined constraints. However, the current MPC methods rely on explicit fluid model of TCP behavior with input time delay. In this paper, we propose a novel AQM algorithm based on data-driven predictive control, called Data-AQM. For Internet system with large delay, complex change and bad disturbance, data-driven predictive controller can be obtained directly based on the input–output data alone and does not require any explicit model of the system. According to the input–output data, the future queue length in data buffer, which is the basis of optimizing drop probability, is predicted. Furthermore, considering system constraints, the control requirement is converted to the optimal control objective, then the drop probability is obtained by solving the optimal problem online. Finally, the performances of Data-AQM are evaluated through a series of simulations.