Assessment of predictive control performance using fractal measures

Abstract

This paper presents novel approach to the task of control performance assessment. Proposed approach does not require any a priori knowledge on process model and uses control error time series data using nonlinear dynamical fractal persistence measures. Notion of the rescaled range R/S plots with estimation of Hurst exponent is applied. Crossover phenomenon is observed in data being investigated and discussed. Paper starts with industrial engineering rationale. Review of the control error histogram is followed by statistical analysis of probabilistic distribution functions (PDFs). Lévy $$\alpha $$ -stable PDF parameters seem to be best fitted. They directly lead to the fractal analysis using Hurst exponents and R/S plot crossover points. The evaluation aims at performance of the generalized predictive control (GPC) and discusses freshly introduced loop performance quality sensitivity against design parameters of the GPC controller.