Using adaptive polynomial models of time series for short-term prediction of the technical parameter

Abstract

The possibilities of using time series for predicting changes in a technical parameter in real time are considered. Forecasting is carried out using simple adaptive models. The prediction procedure should be performed in the background in the microcontroller. Selected adaptive polynomial models of zero, first and second order, based on the method of multiple exponential smoothing. Models are modified to the features of the computation process in the microcontroller. As initial data, the values of the acceleration of the car were used. The forecast was carried out for one or more steps of information retrieval from the accelerometer. The data before the simulation was not pre-processed. Emissions were excluded from the data set. Simulation has shown that the adaptive zero order polynomial model is generally more preferable for one-step prediction. With an increase in the forecasting horizon, the best results are shown by a second-order model.