Non-linear quadratic gaussian control† (Q3318673)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 3849120
Language Label Description Also known as
default for all languages
No label defined
    English
    Non-linear quadratic gaussian control†
    scientific article; zbMATH DE number 3849120

      Statements

      Non-linear quadratic gaussian control† (English)
      0 references
      0 references
      1984
      0 references
      statistical linearization
      0 references
      linear quadratic Gaussian control
      0 references
      nonlinear filtering
      0 references
      nonlinear optimal control
      0 references
      partial information
      0 references
      This paper deals with the optimal steady-state control of nonlinear stochastic systems with a quadratic index, where the controller is chosen to be linear. The nonlinearities are ''statistically linearized'' via multivariate regression on the state variable and at each instant of time a Gaussian distribution is assumed for the state, allowing to obtain ordinary differential equations for the approximate evolution of mean and covariance. Firstly, a complete state information problem is examined for which the Lagrangian equations are turned into Riccati ones, the Gaussian assumption allows to express nonlinearities in functions of the steady- state covariance. Then, the author introduces a linear filter for a nonlinear system obtained by statistical linearization with an optimal choice of the gain matrix. Finally, the case of partial information on the state is examined: the controller is prescribed to be linear in the state estimate whose evolution is prescribed as before, except for the gain. It is observed that a partial separation result holds in that one first solves the Riccati and Ljapunov (steady-state) equations for the filter gain which enter as a parameter in the equations for the control gain. The evaluation of the approximation errors involved is made in some numerical examples.
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references