Nonparametric Techniques in System Identification
This entry gives an overview of classical and state-of-the-art nonparametric time and frequency-domain techniques. In opposition to parametric methods, these techniques require no detailed structural information to get insight into the dynamic behavior of complex systems. Therefore, nonparametric methods are used in system identification to get an initial idea of the model complexity and for model validation purposes (e.g., detection of unmodeled dynamics). Their drawback is the increased variability compared with the parametric estimates. Although the main focus of this entry is on the classical identification framework (estimation of dynamical systems operating in open loop from known input, noisy output observations), the reader will also learn more about (i) the connection between transient and leakage errors, (ii) the estimation of dynamical systems operating in closed loop, (iii) the estimation in the presence of input noise, and (iv) the influence of nonlinear distortions on the linear framework. All results are valid for discrete- and continuous-time systems. The entry concludes with some user choices and practical guidelines for setting up a system identification experiment and choosing an appropriate estimation method.
KeywordsBest linear approximation Correlation method Empirical transfer function estimate Errors-in-variables Feedback Frequency response function Gaussian process regression Impulse transient response modeling method Local polynomial method Local rational method Noise (co)variances Noise power spectrum Spectral analysis
This work is sponsored by the Research Foundation Flanders (FWO-Vlaanderen), the Flemish Government (Methusalem Fund, METH1), the Belgian Federal Government (Interuniversity Attraction Poles programme IAP VII, DYSCO), and the European Research Council (ERC Advanced Grant SNLSID).
- Bendat JS, Piersol AG (1980) Engineering applications of correlations and spectral analysis. Wiley, New YorkGoogle Scholar
- Brillinger DR (1981) Time series: data analysis and theory. McGraw-Hill, New YorkGoogle Scholar
- Carter GC, Nuttall AH (1980) On the weighted overlapped segment averaging method for power spectral estimation. Proc IEEE 68(10):1352–1353Google Scholar
- Eykhoff P (1974) System identification. Wiley, New YorkGoogle Scholar
- Godfrey K (1993) Perturbation signals for system identification. Prentice-Hall, Englewood CliffsGoogle Scholar
- Hägg P, Hjalmarsson H (2012) Non-parametric frequency response function estimation using transient impulse response modelling. Paper presented at the 16th IFAC symposium on system identification, Brussels, 11–13 July, pp 43–48Google Scholar
- Ljung L (1999) System identification: theory for the user, 2nd edn. Prentice-Hall, Upper Saddle RiverGoogle Scholar
- McKelvey T, Guérin G (2012) Non-parametric frequency response estimation using a local rational model. Paper presented at the 16th IFAC symposium on system identification, Brussels, 11–13 July, pp 49–54Google Scholar
- Pintelon R, Schoukens J (2012) System identification: a frequency domain approach, 2nd edn. IEEE, Piscataway/Wiley, HobokenGoogle Scholar
- Stenman A, Gustafsson F, Rivera DE, Ljung L, McKelvey T (2000) On adaptive smoothing of empirical transfer function estimates. Control Eng Pract 8(11):1309–1315Google Scholar