System identification models' fit using error histogram analysis and the Hampel filter as computational tools

In the present investigation, we use the error histogram analysis as a computational tool to define whether the model resulting from a system identification process should continue to be fitted, and the Hampel filter for the elimination of outliers as a tool that also avoids on model over-parameterization. To do this, we use three data sets from a four-cylinder BMW diesel engine, to identify a linear model, and then, with that model, analyze the error and its histogram in a data set (without noise, with noise and with outliers). The analysis of the histogram of the error was found to be a useful tool for detecting white noise and helps to avoid overfitting, in addition to the fact that the Hampel filter allowed detecting and eliminating atypical samples. The software used was MATLAB.