Brief paperStationary behavior of an anti-windup scheme for recursive parameter estimation under lack of excitation☆
Introduction
Consider the following regressor modelwhere is the scalar output measured at discrete time instances , is the regressor vector, is the parameter vector to be estimated and the scalar e is disturbance.
The estimation of is often performed by a linear recursive algorithm of the “prediction–correction” formwhere the first (prediction) term in the right-hand part of the equation highlights the fact that the parameter vector is assumed to be constant.
If is white and the parameter vector is subject to the random walk model driven by a zero-mean white noise sequence the optimal, in the sense of minimum of the a posteriori parameter error covariance matrix, estimate is yielded by (2) with the Kalman gain where , is the solution to the Riccati equationfor some describing the covariance of the initial guess . Optimality of the estimate is guaranteed only whensee Ljung and Gunnarsson (1990). Since these quantities are seldom a priori known, they are usually treated as design parameters of the estimation algorithm and chosen as some in order to achieve desired properties of the filter.
The regressor vector is called persistently exciting (Söderström & Stoica, 1989), if there exist a constant and an integer such that for all tThus, when is persistently exciting, the space is spanned by in at most m steps.
Excitation properties of the regressor vector sequence play an important role in the dynamic behavior of (4). When the excitation in the input data is non-persistent, a phenomenon referred to as (covariance) windup (a.k.a. blow-up) can occur. This means that some eigenvalues of P rise linearly with time.
As pointed out in Cao and Schwartz (2004), the windup phenomenon in the Kalman filter has not been much analyzed until recently. Therefore most of the suggested anti-windup schemes for Kalman filter parameter estimation are of ad hoc nature and are lacking strict proof of non-divergence under lack of excitation, see e.g. Hägglund (1983) and Bittanti, Bolzern, and Campi (1990).
In the approach taken in Stenlund and Gustafsson (2002), which is in the sequel referred to as the Stenlund–Gustafsson (SG) algorithm, a special choice of is used to control the convergence point of the P-matrixwhere . Consequently, the optimality of the Kalman filter estimate is lost.
The structure of (7) is designed to update only in the subspace where excitation is present, that is the image space . Addition of in the denominator of (7) prevents division by zero in case for some t.
A formal proof of the fact that (4), with the free term chosen according to (7), is non-diverging even for the case of lack of excitation, can be found in Medvedev, 2003, Medvedev, 2004. However, stationary properties of the scheme are not considered there.
The SG algorithm can be seen as a generalization of the normalized least mean squares (N-LMS), a method that is well-known and widely used in engineering practice. In Ljung and Gunnarsson (1990), it is shown that the N-LMS can be obtained as a special case of the Kalman filter with the parametersin Eqs. (4), (7). In the N-LMS, the Riccati equation becomes redundant since it is initiated at its stationary point. Thus, the resulting filter (2), (4), (7), (8) is insensitive to loss of excitation. Similarly, in the SG algorithm, once the Riccati equation has converged to the stationary point , it becomes robust against lack of excitation, in the sense that the solution does not diverge.
Interestingly, a directional tracking algorithm presented in Cao and Schwartz (2004), the one identified as Algorithm 1, is also very close to the N-LMS. The suggested choice of the free term in (4) is where and are arbitrary scalars. This algorithm becomes equivalent to the N-LMS with and being initiated at the stationary point of (4), i.e. . Thus, it is also a special case of the SG algorithm.
It is as well worth to note at this point that the proofs of boundedness of the recursive estimation algorithms in Cao and Schwartz (2004) are based on the assumption that all considered solutions to the Riccati equation are positive semidefinite. As mentioned before, this cannot be guaranteed under lack of excitation.
The main result of the paper is formulated in Proposition 4 and provides an explicit parametrization of all stationary solutions to the Riccati equation arising in the SG algorithm. Under sufficient excitation, defined in Section 5, the parametrization implies that the stationary point is unique and is pre-assigned by the matrix . When the excitation is insufficient, the parametrization defines the manifold of all possible solutions, including asymmetric ones.
The paper is organized as follows. First the mechanism behind Riccati equation windup is explained and the problem treated in the article is formulated. Then an equivalent linear time-varying form of the Riccati equation in the SG algorithm (4), (7) is provided. The equation itself was used before, see e.g. Stenlund and Gustafsson (2002), and Medvedev, 2003, Medvedev, 2004, but only a proof for the case when had been originally given in Stenlund and Gustafsson (2002). Then, using the linear form, stationary points of (4), (7) are investigated and some results on the behavior of the algorithm under insufficient excitation are presented.
Section snippets
Riccati equation windup in the Kalman filter
The mechanism behind windup in the Riccati equation can be explained from e.g. random walk model (1), (3). Let the conditions of (5) be used for the Kalman filter design. When is non-singular, all the elements in vary. Note now that (6) can be interpreted as an observability condition of the random walk model at the interval . If (6) does not hold, some of the elements in cannot be observed from the system output y. Since, for the optimal case, describes the covariance of
Problem formulation
In the sequel it is important to distinguish between the stationary solution to the Riccati equation, i.e. when , and stationary data, which means data that have time-invariant statistics.
To address the problem of Riccati equation divergence under lack of excitation, in the SG algorithm is updated only in the excited subspaceThis article deals with the problem of parametrization of stationary
Sylvester equation form
In Stenlund and Gustafsson (2002), it is shown that, for non-singular , the difference obeys the recursionwhere . It immediately follows from (10) that is a stationary point of the difference equation. When excitation is insufficient, positive definiteness of the solution to the Riccati equation cannot be guaranteed. Therefore, before analyzing anti-windup properties of the SG algorithm, it is important to check whether
Stationary points
The purpose of this section is to study stationary points of Eq. (9), arising in the SG algorithm. The stationary solutions are evaluated both for the case when (6) holds and when it does not.
Consider a stationary point of (10)Then the following algebraic condition holds:In vectorized form (14) becomesIn order to separate the direction of excitation at each particular time instant from its intensity, introduce a re-parametrization of the
Conclusion
Stationary properties of a recently suggested windup prevention method for recursive parameter estimation are studied in the case of non-persistently exciting data. A particular choice of the free term of the Riccati equation suggested by the method imposes linear dynamics on the difference Riccati equation and simplifies its analytical analysis.
Generalizing a known result, it is shown that the resulting Riccati equation can always be written as a Sylvester equation. By a direct use of this
Acknowledgements
This work has been in part supported by The Swedish Steel Producers’ Association and by the EC 6th Framework programme as a Specific Targeted Research or Innovation Project (Contract Number NMP2-CT-2003-505467).
Magnus Evestedt was born in Stockholm, Sweden, in 1978. He received a M.Sc. in Engineering Physics from the Uppsala University in 2003. Since 2003 he is a Ph.D. student at the Division of Systems and Control, Department of Information Technology, Uppsala University, Sweden. His research interests include control and monitoring by means of audio and video information.
References (10)
- et al.
Convergence and exponential convergence of identification algorithms with directional forgetting factor
Automatica
(1990) - et al.
Analysis of the Kalman filter based estimation algorithm: An orthogonal decomposition approach
Automatica
(2004) - et al.
Adaptation and tracking in system identification—a survey
Automatica
(1990) New estimation techniques for adaptive control
(1983)- et al.
Topics in matrix analysis
(1991)
Cited by (0)
Magnus Evestedt was born in Stockholm, Sweden, in 1978. He received a M.Sc. in Engineering Physics from the Uppsala University in 2003. Since 2003 he is a Ph.D. student at the Division of Systems and Control, Department of Information Technology, Uppsala University, Sweden. His research interests include control and monitoring by means of audio and video information.
Alexander Medvedev was born in Leningrad (St. Petersburg), USSR (Russia), in 1958. He received his M.Sc (Honors) and Ph.D. degrees in control engineering from Leningrad Electrical Engineering Institute (LEEI) in 1981 and 1987, respectively. From 1981 to 1991, he subsequently kept positions as a system programmer, Assistant Professor, and Associate Professor (docent) in the Department of Automation and Control Science at the LEEI. He was with the Process Control Laboratory, Åbo Akademi, Finland during a long research visit in 1990–1991. In 1991 he has received Docent degree from the LEEI. In the same year, he joined the Computer Science and System Engineering Department at Lulea University of Technology, Sweden as an Associate Professor in the Control Engineering Group (CEG). In February 1996 he has been promoted to Docent.
From October 1996 to December 1997, her served as Acting Professor of Automatic Control in the CEG. Since January 1998, he has been Full Professor of Automatic Control at the same institution.
Starting October 2001, A. Medvedev is also Professor of Automatic Control at Uppsala University, Sweden. He is presently involved in research on fault detection, time-delay systems, time-varying and alternative parameterization methods in analysis and design of dynamic systems. In 1997, in co-operation with a number of leading Swedish process industry companies, he established the Center for Process and System Automation (ProSA) at LTU and until 2002 supervised its activities.
- ☆
This paper was presented at the 16th IFAC World Congress, 4–8 July, 2005, Prague. This paper was recommended for publication in revised form by Associate Editor Antonio Vicino under the direction of Editor Paul Van den Hof.