NIS Test for Selecting the Order of Local Polynomial Model

This work presents an approach that uses normalized innovations squared (NIS) test to determine the order of a local polynomial trend model of a humidity sensor signal. Compared to the classical Akaike information criterion which balances model fit and model complexity NIS test allows to determine the lowest order of the model at which a consistent Bayesian estimation is possible. This approach is useful when signal processing algorithms, such as those for data fusion in multi-sensor systems, contain a Kalman filter, because a successful NIS test is a prerequisite for normal filter operation. The results of the study show that the response of the SHT31-DIS sensor under a step change in humidity is adequately described by a second-order local polynomial model, and this implies a consistent estimation for other input variations.