Model Structure Selection for Speed Forecasting with Nonlinear Autoregressive with an Exogenous Input

This paper compares the model performance for car speed forecasting. The experimented car measures the revolution, injected fuel and current fuel consumption. Nonlinear Autoregressive with an Exogenous Input Model (NARX) and recursive least square (RLS) learning algorithm were selected as a black-box model for forecasting purposes. The input variables were taped from car sensors. The criterions for comparison are based on the mean square error (MSE). Three different inputs for model (NARX1, NARX2 and NARX3) consist of 3000 data collection samples. The first 1500 data were used for training and the rest were used in testing process. The three models (NARX1, NARX2 and NARX3) are selected based on the best performance. The result shows that the model NARX1 outperformed model NARX2 and NARX3 significantly.