| Summary: | The NARX identification process is performed in two steps, namely model structure selection and parameter estimation. Structure selection involves selecting a subset of regressors to use that best describes the system. The structure selection task is a critical part of the system identification process as the number of candidate structures grows exponentially with higher lag spaces. A Binary Particle Swarm based (BPSO) structure selected method has been implemented previously. However, the method was computationally expensive as QR factorization was required for each structure evaluation. Furthermore, redundant structures may be evaluated many times. Because of these problems, a database is proposed to store results of evaluated structures. The proposed method serves as a lookup table to reduce computational time for the BPSO algorithm. This paper describes the implementation of the proposed method, together with analysis of the stored database records on a DC motor dataset. © 2012 IEEE.
|
|---|