A polynomial-algebraic method for non-stationary TARMA signal analysis – Part II: Application to modeling and prediction of power consumption in automobile active suspension systems

Abstract

In this part of the paper, a first application of the non-stationary TARMA approach to the problem of modeling and prediction of a strongly non-stationary mechanical signal is presented, and the approach's suitability, along with the capability of the introduced polynomial-algebraic (P-A) estimation method, are assessed. The signal paradigm used in the study is the power consumption of an automobile hydraulic active suspension system, which is non-stationary due to driving conditions and inherent system characteristics. To render modeling feasible, a general TARMA modeling framework is formulated based upon the P-A estimation method and procedures for subspace dimensionality, basis function, model order selection, and TARMA representation adequacy assessment. These are in turn based upon the notion of model quality-of-fit, which is associated with variable/multiple-step-ahead, instead of the usual one-step-ahead, predictive performance. The results of the study confirm the effectiveness of the approach, revealing the TARMA representation's power and the modeling framework's capabilities in providing accurate signal models. Of particular interest also are critical comparisons, the first of their kind, with previously employed non-stationary ARIMA (integrated ARMA) and conventional ARMA approaches. These reveal the superiority of the TARMA approach in terms of both achievable accuracy and representation simplicity.