Automatica 62 (2015) 184–192
Contents lists available at ScienceDirect
Automatica
journal homepage: www.elsevier.com/locate/automatica
Local optimization of dynamic programs with guaranteed satisfaction
of path constraints
✩
Jun Fu
a
, Johannes M.M. Faust
c
, Benoît Chachuat
b
, Alexander Mitsos
c,1
a
Department of Mechanical Engineering, Massachusetts Institute of Technology (MIT), Cambridge, MA, 02139, USA
b
Centre for Process Systems Engineering, Department of Chemical Engineering, Imperial College, London, SW7 2AZ, UK
c
AVT Process Systems Engineering (SVT), RWTH Aachen University, Turmstrasse 46, 52064 Aachen, Germany
a r t i c l e i n f o
Article history:
Received 15 August 2014
Received in revised form
13 June 2015
Accepted 30 August 2015
Available online 11 November 2015
Keywords:
Dynamic optimization
Path constraints
Semi-infinite programs
Optimization with dynamics embedded
Optimal control
a b s t r a c t
An algorithm is proposed for locating a feasible point satisfying the KKT conditions to a specified tolerance
of feasible inequality-path-constrained dynamic programs (PCDP) within a finite number of iterations.
The algorithm is based on iteratively approximating the PCDP by restricting the right-hand side of the path
constraints and enforcing the path constraints at finitely many time points. The main contribution of this
article is an adaptation of the semi-infinite program (SIP) algorithm proposed in Mitsos (2011) to PCDP.
It is proved that the algorithm terminates finitely with a guaranteed feasible point which satisfies the
first-order KKT conditions of the PCDP to a specified tolerance. The main assumptions are: (i) availability
of a nonlinear program (NLP) local solver that generates a KKT point of the constructed approximation to
PCDP at each iteration if this problem is indeed feasible; (ii) existence of a Slater point of the PCDP that
also satisfies the first-order KKT conditions of the PCDP to a specified tolerance; (iii) all KKT multipliers
are nonnegative and uniformly bounded with respect to all iterations. The performance of the algorithm
is analyzed through two numerical case studies.
© 2015 Elsevier Ltd. All rights reserved.
1. Introduction
Dynamic optimization refers to mathematical programs
whereby the objective and constraint functions depend on the so-
lution of differential or difference equations. Dynamic optimiza-
tion has been widely applied in chemical engineering (Biegler,
2010; Srinivasan, Palanki, & Bonvin, 2003), mechanical engineering
(Hussein & Bloch, 2008; Shin & McKay, 1986), aerospace engineer-
ing (Bainum & Kumar, 1980) and other disciplines (Floudas et al.,
1999). Constrained dynamic optimization problems are practically
important, e.g., to enforce product quality or to guarantee safety
✩
Dr. Jun Fu’s work was partially supported by Natural Sciences and Engineering
Research Council of Canada (NSERC) grant 387509. Prof. Mitsos would like
to acknowledge funding from the DFG-Cluster grant DFG-PAK 635/1 Project
MA 1188/34-1 ‘‘Strategie zur robusten dynamischen Echtzeitoptimierung’’ and
discussions within that cluster. The material in this article was not presented at
any conference. This paper was recommended for publication in revised form by
Editor Berç Rüstem.
E-mail addresses: junfu@mit.edu, fujuncontrol@gmail.com (J. Fu),
johannes.faust@avt.rwth-aachen.de (J.M.M. Faust), b.chachuat@imperial.ac.uk
(B. Chachuat), amitsos@alum.mit.edu (A. Mitsos).
1
Tel.: +49 2418094704; fax: +49 2418092326.
(Feehery & Barton, 1998; Srinivasan et al., 2003). Constraints fall
in either one of two categories, namely point constraints and path
constraints. The former are usually expressed as functions of the
states at the end of time horizon, whereas the latter are functions
of the states and/or controls over the entire time horizon. The fo-
cus of this article is on dynamic optimization with path constraints.
Point constraints, which do not pose any further complication for
the approach herein, are omitted for simplicity. Throughout the ar-
ticle, it is assumed that a control vector parameterization has been
performed, i.e., a finite number of decision variables is assumed.
Numerical solution methods for such dynamic optimization
problems rely on nonlinear programming (NLP) techniques, ei-
ther with or without parameterization of the state trajectories.
In the simultaneous method, also known as orthogonal colloca-
tion approach (Betts & Huffman, 1992; Biegler, 2007; Tsang, Him-
melblau, & Edgar, 1975), the state trajectories are parameterized
and the residuals of the differential equations are enforced as con-
straints at specified collocation times. In the sequential method
(Biegler, 2010; Goh & Teo, 1988), the state trajectories are re-
garded as functions of the control decision variables. In the direct
multiple shooting method (Bock & Plitt, 1984), the state trajecto-
ries are formed by piecing together those of finite single shooting
problems on the corresponding subintervals over which the pa-
rameterized control is applied (see p. 243 in Bock & Plitt, 1984).
http://dx.doi.org/10.1016/j.automatica.2015.09.013
0005-1098/© 2015 Elsevier Ltd. All rights reserved.