Real time neural network learning with lost packets using sliding window approaches

This paper presents real time nonlinear system identification with irregular sampling time or lost packets. This work views the performance of predictive MLP neural network using sliding window learning approach. By adopting nonlinear autoregressive with external input (NARX) model order, this paper investigate the response of sliding window leaning when the measurement received by the MLP network are susceptible to random loss. The simulation results show that the sliding window approach yields good convergence despite the information being lost overtime. The paper concludes that result obtained from sliding window conjugate gradient (with Dai and Yuan variant) has the best convergence rate.