Download E-books Combined Parametric-Nonparametric Identification of Block-Oriented Systems (Lecture Notes in Control and Information Sciences) (Volume 454) PDF

By Grzegorz Mzyk

This publication considers an issue of block-oriented nonlinear dynamic procedure identity within the presence of random disturbances. This type of platforms contains a variety of interconnections of linear dynamic blocks and static nonlinear components, e.g., Hammerstein procedure, Wiener method, Wiener-Hammerstein ("sandwich") procedure and additive NARMAX platforms with suggestions. Interconnecting signs usually are not available for dimension. The mixed parametric-nonparametric algorithms, proposed within the ebook, will be chosen dependently at the past wisdom of the approach and indications. such a lot of them are in response to the decomposition of the advanced process id job into less complicated neighborhood sub-problems by utilizing non-parametric (kernel or orthogonal) regression estimation. within the parametric degree, the generalized least squares or the instrumental variables approach is usually utilized to deal with correlated excitations. restrict houses of the algorithms were proven analytically and illustrated in uncomplicated experiments.

Show description

Read Online or Download Combined Parametric-Nonparametric Identification of Block-Oriented Systems (Lecture Notes in Control and Information Sciences) (Volume 454) PDF

Best System Theory books

Fuzzy Systems Theory and Its Applications

Fuzzy platforms is a mathematical framework for facing doubtful, ambiguous, and approximate details. a substitute for conventional binary common sense, fuzzy common sense has proved to be larger suited for fixing complicated, genuine international programming difficulties. **This ebook offers a concise review of fuzzy structures idea and its purposes.

Hidden Order: How Adaptation Builds Complexity (Helix Books)

Explains how scientists who research complexity are confident that sure consistent procedures are at paintings in every kind of unrelated complicated platforms.

Complexity Metaphors, Models, and Reality

The phrases complexity, complicated adaptive platforms, and sciences of complexity are came upon frequently in fresh clinical literature, reflecting the amazing progress in collaborative educational examine concerned about complexity from the starting place and dynamics of organisms to the most important social and political companies.

Theory and Practice of Recursive Identification (Signal Processing, Optimization, and Control)

Tools of recursive id care for the matter of establishing mathematical types of signs and structures online, even as information is being accumulated. Such tools, that are sometimes called adaptive algorithms or sequential parameter estimation tools, could be utilized to a large spectrum of on-line adaptive platforms, corresponding to units for sign processing, prediction, or keep watch over and are important for modeling structures more often than not.

Extra resources for Combined Parametric-Nonparametric Identification of Block-Oriented Systems (Lecture Notes in Control and Information Sciences) (Volume 454)

Show sample text content

Four we practice nonparametric (kernel or orthogonal) estimate of the regression functionality within the first Hammerstein procedure, which performs the function of deconvolution of the method xk . The received estimates ηM (uk ), computed from M observations {(uk , xk )}M k=1 are plugged in to the vectors of N tools ψk (k = 1, 2, ... , N ) within the parametric id process of moment procedure. The relation among N and M , which promises consistency of the mixed parametric-nonparametric estimate, is additionally given. identified {η(uk )} allow us to begin from the best case, within which the interior sign (nonlinearity output) η(uk ) of the 1st Hammerstein procedure might be saw and permit the linear dynamic block {λi }∞ i=0 is still unknown. the next theorem holds. Theorem 2. eleven. The instrumental variable matrix ΨN generated in accordance with the guideline T Ψ N = (ψ1 , ψ2 , ... , ψN )T , the place ψ okay = 1, η(uk ), η 2 (uk ), ... , η W −1 (uk ) , (2. 112) 74 2 Hammerstein procedure fulfils (C1) and (C2). evidence. For the evidence see Appendix A. thirteen. Unknown {η(uk )} (IV ) −1 T T 1 Theorem 2. eleven promises that the estimate pN = N1 Ψ N ΦN N Ψ N YN is strongly constant, i. e. , converges with likelihood 1 to the parameter vector p of the pointed out functionality µ(x). in spite of the fact that, the method {η(uk )}, showing within the tools ψ ok , can't be measured and its estimation is required. equally as in [98] we advise the initial deconvolution of {xk } to supply estimates ηM (uk ) of η(uk ), yet within the challenge thought of the following, we will practice nonparametric regression estimation tools. The inverse filtering of {xk } is made via the normal kernel or orthogonal algorithms of the regression functionality estimation within the first Hammerstein process, utilizing the pairs {(uk , xk )}M k=1 . We emphasize that ηM (uk )’s are computed utilizing the series of M measurements, which often is bigger than N , utilized in the parametric (instrumental variables) degree. the next theorem exhibits the relation among N and M, which promises the consistency of the two-stage estimate. Theorem 2. 12. If the estimation errors of ηM (uk ) behaves like |ηM (uk ) − η(uk )| = O(M −τ ) with likelihood 1 as M → ∞, (2. 113) then for the tools ΨN,M = (ψ1,M , ψ2,M , ... , ψN,M )T , W −1 2 the place ψk,M = 1, ηM (uk ), ηM (uk ), ... , ηM (uk ) (2. 114) T , the two-stage estimate (IV ) pN,M = 1 T Ψ ΦN N N,M −1 1 T Ψ YN N N,M (2. one hundred fifteen) is strongly constant, i. e. , (IV ) pN,M → p with likelihood 1 as N, M → ∞, only if N M −τ → zero. facts. From (2. 113) we finish that for all w = 1, 2, ... , W and ok = w 1, 2, ... , N , it holds that |ηM (uk ) − η w (uk )| = O(M −wτ ), and for that reason ψk,M − ψ okay = O(M −wτ ) with likelihood 1, as M → ∞, the place ψ ok is given through (2. 112). Now, Theorem 2. 12 should be proved in an analogous demeanour 2. five Cascade platforms seventy five as Theorem three in [65], with noticeable substitutions, and accordingly the evidence is right here passed over. The comment less than, supplies optimistic advice for the choice of τ . comment 2. 15. [50] If η() and the enter chance density f () are at the least twice always differentiable at u, then for hM = h0 M −1/5 the convergence cost is |ηM (u) − η(u)| = O(M −2/5 ) in chance, i.

Rated 4.74 of 5 – based on 35 votes