Two new square root Kalman filtering algorithms are presented. Both algorithms are based on the spectral V − Λ of the covariance matrix where V is the matrix whose columns are the eigenvectors of the covariance and ...
详细信息
Two new square root Kalman filtering algorithms are presented. Both algorithms are based on the spectral V − Λ of the covariance matrix where V is the matrix whose columns are the eigenvectors of the covariance and Λ is the diagonal matrix of its eigenvalues. The algorithms use the covariance mode in the time propagation stage and the information mode in the measurement update stage. This switch between modes, which is trivial in the V − Λ representation, increases the efficiency of the algorithms. In the first algorithm, which is a continuous/discrete one, the V and Λ 1 2 matrices are propagated in time in a continuous manner, while the measurement update is a discrete time procedure. In the second algorithm, which is a discrete/discrete one, the time propagation of the V − Λ 1 2 factors is performed in discrete time too, using a procedure which is similar to the one used for the discrete measurement update. The discrete propagation and the measurement update are based on singular value decomposition algorithms. The square root nature of the algorithms is demonstrated numerically through a typical example. While promising all the virtues of square root routines, the V − Λ filters are also characterized by their ability to exhibit singularities as they occur.
作者:
MOORE, JBBOEL, RKDepartment of Systems Engineering
Research School of Physical Sciences Australian National University Canberra ACT 2601 Australia Visiting Research Fellow (NFWO)
Department of Systems Engineering Research School of Physical Sciences Australian National University on leave from an NFWO (Belgium National Foundation for Scientific Research) Research Fellow position at the Rijksuniversiteit Gent Belgium
The challenge taken up in this paper is to devise a parameter identification algorithm for linear, discrete-time, stochastic plants which exploits the strengths of both the extended least squares (ELS) and the recursi...
详细信息
The challenge taken up in this paper is to devise a parameter identification algorithm for linear, discrete-time, stochastic plants which exploits the strengths of both the extended least squares (ELS) and the recursive prediction error (RPE) parameter estimation methods. The focus is on adaptive control of parameterized state space models which exploit a priori plant information in that the unknown parameter vector θ r is of lower dimension than that for a corresponding input-output model parameterized by θ. A triple parameter estimation scheme consisting of ELS, RPE and a hybrid of the two, denoted HPE, is proposed. The purpose of the HPE scheme is to permit information flow from the ELS to RPE algorithms so as to effectively project RPE into a stability domain, and to have it avoid local prediction error index minima that are not the global minimum.
作者:
DUPUIS, PKUSHNER, HJBrown Univ
Div of Applied Mathematics Providence RI USA Brown Univ Div of Applied Mathematics Providence RI USA
Asymptotic properties of Robbins–Munro and Kiefer–Wolfowitz type stochastic approximation algorithms are obtained via the theory of large deviations. The conditions are weak and can even yield w.p.l. convergence res...
详细信息
Asymptotic properties of Robbins–Munro and Kiefer–Wolfowitz type stochastic approximation algorithms are obtained via the theory of large deviations. The conditions are weak and can even yield w.p.l. convergence results. The probability of escape of the iterates from a neighborhood of a stable point of the algorithm is estimated and shown to be considerably smaller than suggested by the classical “asymptotic normality of local normalized errors” method of getting the asymptotic properties. The escape probabilities are a natural quantity of interest. In many applications, they are more useful than the “local normalized mean square errors.” Other large deviations estimates are also obtained. Typically, if ${{a_n = 1} / {n^\rho }}$, $\rho \leqq 1$, then the probability of escape from a neighborhood of a stable point in some (normalized) time interval $[n,m]:\sum_n^m {a_i \sim T} $ is $\exp - n^\rho V_\rho $, where $V_\rho $ does not depend on $\rho $ for $\rho < 1$ and is the solution to an optimal control problem. If the noise is Gaussian, then the optimal control problem is relatively easy. Under quite broad conditions, in the Kiefer–Wolfowitz case the control problem has the Gaussian form, whether or not the noise is Gaussian. The techniques are expected to be quite useful in the analysis of the asymptotic properties of recursive algorithms generally.
作者:
EWEDA, EMACCHI, OCNRS
ECOLE SUPER ELECTSIGNAUX & SYST LABF-91190 GIF SUR YVETTEFRANCE
Adaptive filtering with error gradient algorithm and constant step-size is analyzed for a deterministic time variable optimum filtering vector. The unrealistic assumption of independent observations is replaced by a b...
详细信息
Adaptive filtering with error gradient algorithm and constant step-size is analyzed for a deterministic time variable optimum filtering vector. The unrealistic assumption of independent observations is replaced by a bounded memory model, largely justifiable in applications. Then the mean square tracking deviation (MSD) between the optimum vector and the algorithm output is proved to include two contributions; the stationary mode error, characteristic of convergence accuracy, which is proportional to the step-size; and the transient mode error, reflecting the rapidity of tracking, which is proportional to the squared ratio of the maximum optimum estimator increment to the step-size. This result agrees with the common intuition that there exists an optimum step-size which compromises between convergence accuracy and tracking speed.
A general problem involving optimization of a covariance sequence is considered in the paper. One difficulty with this class of problems is to ensure that the covariance sequence is nonnegative definite (in other word...
详细信息
A general problem involving optimization of a covariance sequence is considered in the paper. One difficulty with this class of problems is to ensure that the covariance sequence is nonnegative definite (in other words, realizable). It is suggested that this difficulty can be overcome by reformulating the optimization problem in terms of the partial autocorrelation coefficients (PAC). One need only constrain these coefficients to lie in the range (−1, 1) to guarantee that the corresponding covariance sequence is nonnegative definite. The synthesis of a signal realizing the optimizing covariance sequence is also discussed. Special emphasis is given to the case when some of the PACs are either +1 or −1.
Monte Carlo simulation is used to compare the numerical performance of a second-order filtering algorithm with those of two previously tested filters in a well-known nonlinear state estimation problem.
Monte Carlo simulation is used to compare the numerical performance of a second-order filtering algorithm with those of two previously tested filters in a well-known nonlinear state estimation problem.
Computing the control law for non-linear systems can be facilitated by using a stochastic automaton model. The automaton adapts i5 policy by means of a stochastic gradient algorithm, that depends on the estimation of ...
详细信息
Computing the control law for non-linear systems can be facilitated by using a stochastic automaton model. The automaton adapts i5 policy by means of a stochastic gradient algorithm, that depends on the estimation of the parameters. Numerical simulations concerning the control of a non-linear process are presented.
The problem of controlling a plant with unknown and slowly time-varying para meters is considered. In order to make the resulting adaptive controller robust against model mismatching, future bahaviour of the plant is ...
详细信息
The problem of controlling a plant with unknown and slowly time-varying para meters is considered. In order to make the resulting adaptive controller robust against model mismatching, future bahaviour of the plant is not described by a single predictive model but by a number of different predictive models whose parameters are independently estimated. Attention is devoted to developing recursive least-squares (RLS) estimators consisting of UD covariance filters or square root information filters that, by exploiting the relationship among different regressors, keep the overall numerical burden as small as possible
The problem of adaptively controlling a linear multivariable plant according to a quadratic cost functional defined over a control horizon of arbitrary length is discussed. In this context, the proposed algorithm, ref...
详细信息
The problem of adaptively controlling a linear multivariable plant according to a quadratic cost functional defined over a control horizon of arbitrary length is discussed. In this context, the proposed algorithm, referred to by the acronym MUSMAR, is shown to be a natural generalization of standard self-tuning controllers. By increasing the control horizon length, the MUSMAR closely approximates a steady-state LQG controller inheriting the intrinsic robustness of LQG design. Analysis and simulations give evidence of several attractive features of the MUSMAR self-tuner when applied to plants for which standard adaptive controllers fail to converge or yield an unacceptable performance.
This paper presents an overview of the current status of convergence theory for adaptive control algorithms. Rather than giving a comprehensive survey, the paper aims to emphasize the conceptual common ground between ...
详细信息
This paper presents an overview of the current status of convergence theory for adaptive control algorithms. Rather than giving a comprehensive survey, the paper aims to emphasize the conceptual common ground between different approaches. Possible areas for future research are also discussed.
暂无评论