Long-memory modeling and prediction of automotive active suspension power consumption

Abstract

This paper considers long-memory modeling and prediction of power consumption in automotive fully-active suspension systems. The study is based upon a novel pseudo-linear method for the estimation of long-memory fractionally integrated ARMA (ARFIMA) models and an experimental active suspension vehicle. In addition to nonstationarity, the power consumption signal is shown to possess long-memory characteristics which are effectively captured by the ARFIMA model structure. Its comparison with alternative model structures indicates its superiority over conventional ARMA/ARIMA approaches. Although its achieved predictive accuracy is somewhat lower than that of fundamentally nonstationary time-dependent ARMA (TARMA) models, the ARFIMA model is shown to achieve satisfactory performance at a drastically reduced parametric complexity, which is lowest among all considered model structures.