A Time-varying Neural-network-based ARX Model of the Traffic in Mobile Networks

This paper presents a time-varying discrete-time neural-network-based auto-regressive model with exogenous inputs (NN-ARX), which is continuously updated using the online sequential extreme learning algorithm (OS-ELM). The model exhibited a consistently satisfactory performance when describing the dynamics of the payload as a function of users, from real trac data of two stations in a 4G-LTE network. Some advantages of the proposed NN-ARX model include: (i) it provides more modeling flexibilty than classical nonlinear-regression-approaches, (ii) it has adaptive capabilities to characterize the time-varying-behavior of dierent cells and (iii) it has low computational cost, which makes the model suitable for the implementation of optimization and control strategies to improve the performance of the network.