Equi-normalization and exact scaling dynamics in homothetic tube model predictive control
Introduction
Model predictive control (MPC) need not be robust [1], [2], [3], while min–max feedback MPC [4] is computationally intractable, hence the need for robust MPC synthesis that strikes a balance between optimality and computability. Tube MPC (TMPC) [2], [3], [5], [6] forms a rigorous set theoretic approach to robust MPC synthesis, which results in computationally efficient handling of constraints and uncertainty. Early TMPC [3], [5] used “rigid” tubes with fixed cross-section shape sets whose centers were optimized on-line. The deployment of the fixed tube cross-section shape sets results in a degree of conservatism. Latter “homothetic” tubes allowed for more flexibility: the cross-sections of homothetic state and control tubes are parameterized in terms of the sequences of associated centers and scalings [7]. These sequences form decision variables for the on-line optimization, and are subject to constraints that enforce robust constraint satisfaction. Global conditions describing the tube dynamics and ensuring robust constraint satisfaction were given in [7]. The dynamics of the local homothetic state and control tubes is characterized by the dynamics of the local state and control tubes centers and scalings. These dynamics plays a crucial role for derivation of the stabilizing conditions through the use of a terminal constraint set and a local Lyapunov function. In [7], an affine relationship between the tube scalings at consecutive times was used. The affine scalings dynamics represents an upper approximation of the exact dynamics and, as such, it induces a degree of conservatism.
This article introduces an equi-normalization process, derives a description of the exact scaling dynamics and achieves a significant constraint relaxation. Using the proposed equi-normalization we establish the existence of, and characterize, the minimal robust positively invariant (RPI) set generated by a number of a-priori, but suitably, selected linear inequalities. The function inducing the exact scaling dynamics is shown to be a piecewise affine, convex, continuous, non-negative and monotonically non-decreasing function. Remarkably, this function is composed of, at most, two affine functions irrespective of the system dimension and the number of utilized linear inequalities. In particular, this function can be expressed as the maximum of two affine functions and yields the exact scaling dynamics (in contrast to the affine upper approximate scaling dynamics of [7]). The equi-normalized minimal RPI (mRPI) set together with the exact scaling function is used to analyze the dynamics of the local state and control tubes centers and scalings . (The latter dynamics is determined completely by the -dynamics.) This analysis enables us to ensure the maximality (w.r.t. both the volume and set inclusion) of the constraint set on the local state and control tubes centers and scalings . (These constraints are characterized fully by constraints on the -variable.) The equi-normalized mRPI set induces the minimal (w.r.t. set inclusion) homothetic tubes cross-section shape sets, while the exact scaling function induces the tightest (w.r.t. set inclusion) -dynamics. In turn, this yields a significantly improved terminal constraint set, and relaxed conditions for the applicability of both the homothetic tube optimal control (HTOC) and homothetic tube MPC (HTMPC). It is also shown that the equi-normalized mRPI set, the exact scaling dynamics and the enlarged terminal constraint set can be incorporated into HTMPC through a generalization of the cost and terminal cost functions and a local Lyapunov condition. This leads to an improved HTMPC synthesis, which is applicable under relaxed conditions compared to the existing approaches [5], [7].
Paper structure: Preliminaries and brief overview of the existing TMPC syntheses are given in Section 2, while Section 3 introduces equi-normalization, and derives the exact tube scaling dynamics and improved terminal constraint set. The improved HTOC and HTMPC are outlined in Section 4. Comparative remarks are presented in Section 5, while conclusions are drawn in Section 6.
Nomenclature and typographical conventions: The sets of non-negative and positive integers and non-negative reals are denoted by , , and , respectively. Given with we denote ; we write for . For , inequalities such as apply component-wise. For , denotes its spectral radius. Given and , the Minkowski set addition is defined by , we write instead of . Given a set and a real matrix of compatible dimensions, the image of under is denoted by . A function is said to be positively homogeneous of the first degree if for all and all . If is a set-valued function from, say, into , then its graph is the set . A set is a -set if it is compact, convex, and contains the origin; if in addition the origin is in the interior of , then is said to be a proper -set, or just a -set. A polyhedron is the (convex) intersection of a finite number of open and/or closed half-spaces and a polytope is a closed and bounded polyhedron. We say that a set is a (-) -polytopic set if it is a polytope which is a (-) -set. denotes the interior of a set whereas denotes its convex hull. Sets and are homothetic if and only if for some and . Given a non-empty closed convex set the associated support function is given, for all , by . We distinguish the row and column vectors only when needed, we use the same symbol for a variable and its vectorized form, and we provide brief proofs of less obvious statements in the Appendix.
Section snippets
Preliminaries
Consider linear, time-invariant, discrete time systems and constraints: The variables and are, respectively, the current state, disturbance and control, while is the successor state. Clearly, at any time instance , system (2.1) reads as .
Assumption 1 The matrix pair is stabilizable and is known exactly as is when the current control action is determined. The current and future disturbances are not known but satisfy
Equi-normalization and exact homothetic tube scaling dynamics
Assumption 3(ii) implies that the set admits an irreducible representation: is a finite integer, each , and spans . Thus, for all , the set is guaranteed to be at least a -polytopic set in . Note that the irreducibility of the representation of the set means that none of the inequalities is redundant.
The focus of this section is on the local tubes centers and scalings -dynamics
Improved HTOC and HTMPC
The deployment of the equi-normalized mRPI set , the exact scaling dynamics and the improved terminal constraint set results in HTMCP synthesis that, under relaxed assumptions, outperforms RTMPC of [5] and HTMPC of [7]. Our developments mirror those reported in [7] albeit requiring some distinct modifications; thus, our discussion is brief and focuses on the required modifications. We consider the homothetic state and control tubes and as specified in (2.6), (2.7), (2.8),
Computational and comparative remarks
As in the case of [5], [7], the main computational burden of the improved HTMPC synthesis is the off-line determination of , , , and . The interested reader is referred to [7, Section 6] for a more detailed discussion.
In contrast to the simple HTMPC of [7], the improved proposal requires an additional step (the equi-normalization of the initial RPI set in order to obtain the set ) in the off-line design. This task can be accomplished with relative ease, since for a given ,
Conclusions
We have developed the equi-normalization process and have verified the existence and obtained the characterization of the mRPI set generated by a number of a-priori, but suitably, selected linear inequalities. We have also derived the exact scaling dynamics associated with this mRPI set. We have utilized the mRPI set and the exact scaling dynamics to improve the existing TMPC methods and relax conditions for their applicability. It has been shown that the use of equi-normalization and exact
Acknowledgments
The authors are grateful to the Editor, associate editor and reviewers for constructive comments. This paper has been finalized while Saša V. Raković was a Visiting Assistant Professor at the Institute for Systems Research of the University of Maryland. His visit and work were supported by the University of Maryland and the Foundation of the University of Maryland.
References (13)
- et al.
Examples when nonlinear model predictive control is nonrobust
Automatica
(2004) - et al.
Robust model predictive control of constrained linear systems with bounded disturbances
Automatica
(2005) - et al.
Homothetic tube model predictive control
Automatica
(2012) Set theoretic methods in model predictive control
Lecture Notes in Control and Information Sciences—Nonlinear Model Predictive Control : Towards New Challenging Applications
(2009)- et al.
Model Predictive Control: Theory and Design
(2009) - et al.
Min–max feedback model predictive control for constrained linear systems
IEEE Transactions on Automatic Control
(1998)
Cited by (32)
Configuration-Constrained Tube MPC
2024, AutomaticaEllipsoidal approximations of the minimal robust positively invariant set
2023, ISA TransactionsRobust coalitional model predictive control with negotiation of mutual interactions
2023, Journal of Process ControlA novel dual-mode robust model predictive control approach via alternating optimizations
2021, AutomaticaCitation Excerpt :For example, an outer-approximation to the mRPI set of a linear autonomous system was obtained by solving a sequence of linear programs (LPs), and computing the Minkowski summations (Rakovic et al., 2005). Among several suitably selected linear inequalities, one will be able to determine the polytopic mRPI set by solving an iterative LP sequence (Raković et al., 2013). The results were further improved by considering solving a single LP (Trodden, 2016).
A set-theoretic generalization of dissipativity with applications in Tube MPC
2020, AutomaticaCitation Excerpt :As rigid tubes may be rather conservative, multiple strategies have been proposed to increase the accuracy of RFITs. These include the use of homothetic (Raković, Kouvaritakis, & Cannon, 2013; Raković, Kouvaritakis, Findeisen & Cannon, 2012) and elastic tube parameterizations (Raković, Levine, & Açıkmeşe, 2016), which are based on polytopic sets with a constant, pre-specified number of facets. The use of ellipsoidal parameterizations, Villanueva, Quirynen, Diehl, Chachuat, and Houska (2017), has also been proposed for tube MPC.