Extended analysis of bpso structure selection of nonlinear auto-regressive model with exogenous inputs ( NARX) of direct current motor

Abstract

System Identification (SI) is a discipline concerned with inference of mathematical models from dynamic systems based on their input and output measurements. Among the many types of SI models, the superior NARMAX model and its derivatives (NARX and NARMA) are powerful, efficient and unified representations of a variety of nonlinear systems. The identification process of NARX/NARMA/NARMAX is typically performed using the established Orthogonal Least Squares (OLS). Weaknesses of the OLS model are known, leading to various alternatives and modifications of the original algorithm. This paper extends the findings of previous research in application of the Binary Particle Swarm Optimization (BPSO) for structure selection of a polynomial NARX model on a DC Motor (DCM) dataset. The contributions of this paper involve the implementation and analysis of a MySQL database to serve as a lookup table for the BPSO optimization process. Additional analysis regarding the frequencies of term selection is also made possible by the database. An analysis of different preprocessing methods was also performed leading to the best model. The results show that the BPSO structure selection method is improved by the presence of the database, while the magnitude scaling approach was the best preprocessing method for NARX identification of the DCM dataset.

Keywords

• nonlinear system identification
• nonlinear autoregressive with exogenous inputs model (narx)
• structure selection
• binary particle swarm optimization algorithm (BPSO)
• direct current motor