Nonlinear Filters. Simon Haykin

Читать онлайн книгу.

Nonlinear Filters - Simon  Haykin


Скачать книгу
system for the reconstruction error in bold z:

      (3.17)bold e Subscript k minus n Sub Subscript x Subscript plus 2 vertical-bar k plus 1 Baseline equals left-parenthesis bold upper A minus bold upper L bold upper C right-parenthesis bold e Subscript k minus n Sub Subscript x Subscript plus 1 vertical-bar k Baseline plus upper O left-parenthesis parallel-to bold e Subscript k minus n Sub Subscript x Subscript plus 1 vertical-bar k Baseline parallel-to right-parenthesis comma

      where bold upper C equals Start 1 By 4 Matrix 1st Row 1st Column 0 2nd Column midline-horizontal-ellipsis 3rd Column 0 4th Column 1 EndMatrix and

bold upper A equals Start 5 By 5 Matrix 1st Row 1st Column 0 2nd Column 0 3rd Column midline-horizontal-ellipsis 4th Column 0 5th Column 0 2nd Row 1st Column 1 2nd Column 0 3rd Column midline-horizontal-ellipsis 4th Column 0 5th Column 0 3rd Row 1st Column 0 2nd Column 1 3rd Column midline-horizontal-ellipsis 4th Column 0 5th Column 0 4th Row 1st Column vertical-ellipsis 2nd Column vertical-ellipsis 3rd Column down-right-diagonal-ellipsis 4th Column vertical-ellipsis 5th Column vertical-ellipsis 5th Row 1st Column 0 2nd Column 0 3rd Column midline-horizontal-ellipsis 4th Column 1 5th Column 0 EndMatrix period

      This approach for designing nonlinear observers is applicable to systems that are observable for every bounded input, parallel-to bold u Subscript k Baseline parallel-to less-than-or-equal-to upper M, with bold g left-parenthesis bold x comma period right-parenthesis and bold upper F Superscript n Baseline left-parenthesis bold upper Phi Superscript negative 1 Baseline left-parenthesis bold x comma period right-parenthesis right-parenthesis being uniformly Lipschitz continuous functions of the state:

      (3.18)sup Underscript parallel-to bold u Subscript k Baseline parallel-to less-than-or-equal-to upper M Endscripts parallel-to bold g left-parenthesis bold x 1 comma period right-parenthesis minus bold g left-parenthesis bold x 2 comma period right-parenthesis parallel-to less-than-or-equal-to upper L 1 parallel-to bold x 1 minus bold x 2 parallel-to comma

      (3.19)sup Underscript parallel-to bold u Subscript k Baseline parallel-to less-than-or-equal-to upper M Endscripts parallel-to bold upper F Superscript n Baseline left-parenthesis bold upper Phi Superscript negative 1 Baseline left-parenthesis bold x 1 comma period right-parenthesis right-parenthesis minus bold upper F Superscript n Baseline left-parenthesis bold upper Phi Superscript negative 1 Baseline left-parenthesis bold x 2 comma period right-parenthesis right-parenthesis parallel-to less-than-or-equal-to upper L 2 parallel-to bold x 1 minus bold x 2 parallel-to comma

      where upper L 1 and upper L 2 denote the corresponding Lipschitz constants. However, convergence is guaranteed only for a neighborhood around the true state [36].

      The equivalent control approach allows for designing the discrete‐time sliding‐mode realization of a reduced‐order asymptotic observer [37]. Let us consider the following discrete‐time state‐space model:

      (3.22)bold upper T bold x Subscript k Baseline equals StartBinomialOrMatrix bold z Subscript k Superscript u Baseline Choose bold z Subscript k Superscript l Baseline EndBinomialOrMatrix comma

      such that the upper partition has the identity relationship with the output vector:

      (3.23)bold y Subscript k Baseline equals bold z Subscript k Superscript u Baseline period

      where

      (3.25)bold upper Phi equals bold upper T bold upper A bold upper T Superscript negative 1 Baseline equals Start 2 By 2 Matrix 1st Row 1st Column bold upper Phi 11 2nd Column bold upper Phi 12 2nd Row 1st Column bold upper Phi 21 2nd Column bold upper Phi 22 EndMatrix comma

      (3.26)bold upper G equals bold upper T bold upper B equals StartBinomialOrMatrix bold upper G 1 Choose bold upper G 2 EndBinomialOrMatrix period

      The corresponding sliding‐mode observer is obtained as: