Classification of Distributions of the Values of Impulses Forcing Vibrations of an Oscillator

Abstract

The present work discusses the classification of distribution of values of impulses. We solve the problem in which random impulses occur before the vibrations caused by previous random impulses expire. The goal was to achieve a correct classification of three groups of distributions characteristic of three different processes: correct, transitory and flawed. The starting moment of the analysis as well as the length of time interval are the objects of study of this article. In order to determine the significance of the extracted features of the random time series under study, SHapley Additive exPlanations method was applied. The paper discussed the stage of study which is focused on finding the first three stochastic raw moments that allow for the most precise classification of distributions characterizing different processes in the shortest time. The analysis using the random forest classifier made it possible to distinguish the distributions characteristic for different types of processes, that is, it is possible to distinguish correct processes and transitory ones, etc. It is hardest to distinguish correct processes from one another, since they are highly similar to one another, but different from the other types.

Keywords

• Stochastic mechanics
• random series of impulses
• stationary time series
• non-stationary time series
• classification
• feature engineering