Multiple Shooting and Time Domain Decomposition May. 6-8, 2013, IWR, Heidelberg, Germany Multiple-Boundary-Value-Problem Formulation for PDE constrained Optimal Control Problems with a Short History on Multiple Shooting for ODEs Hans Josef Pesch Chair of Mathematics in Engineering Sciences University of Bayreuth, Germany [email protected]
45
Embed
Multiple-Boundary-Value-Problem Formulation for PDE constrained Optimal Control Problems
Multiple-Boundary-Value-Problem Formulation for PDE constrained Optimal Control Problems with a Short History on Multiple Shooting for ODEs Hans Josef Pesch Chair of Mathematics in Engineering Sciences University of Bayreuth, Germany [email protected]. Outline. - PowerPoint PPT Presentation
Welcome message from author
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Multiple-Boundary-Value-Problem Formulationfor PDE constrained Optimal Control Problems
with a Short History on Multiple Shooting for ODEs
Hans Josef PeschChair of Mathematics in Engineering Sciences
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Outline
• A short history on multiple shooting
• Multipoint-boundary-value-problem formulation
• A state constrained elliptic problem
• A state constrained parabolic PDE-ODE problem • A singular hyperbolic optimal control problem
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Outline
• A short history on multiple shooting
• Multipoint-boundary-value-problem formulation
• A state constrained elliptic problem
• A state constrained parabolic PDE-ODE problem • A singular hyperbolic optimal control problem
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
The not Well-known Stone Age of Multiple Shooting
Engineers: Morrison, Riley, Zancanaro (1962)
Multiple Shooting Method for Two-Point Boundary Value Problems,Communications of the ACM, 1962, pp. 613 - 614.
One serious shortcoming of shooting becomes apparent when, as happens altogether too often, the differential equations are so unstablethat they „blow up“ before the initial value problem can be completely integrated.This can occur even in the face of extremely accurate guesses for the initial values. Hence, shooting seems to offer no hope for some problems. A finite difference method does have a chance for it tends to keep a firm hold on the entire solutionat once. The purpose of this note is to point out a compromising procedurewhich endows shooting-type methods with this particular advantage of finite difference methods. For such problems, then, all hope need not be abandoned for shooting methods. This is desirable because shooting methods are generally faster than finite difference methods.
Parallel shooting on equidistant intervals
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
The Pioneers
Keller, Osborne (1968,69): first analysis
Bulirsch, Stoer (1971,73): first algorithmic realisation
Concept and first analysis of multiple shooting and parallel shooting
First code (1968): BOUNDSOL: nonlinear boundary value problemsSecond code (1970): OPTSOL: optimal control problems with inequality constraints
Bulirsch coined the term Mehrzielmethode
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
The Followers
Deuflhard (1974,75): improved Newton method (DLOPTR)
Oberle (1977,83): multipoint bvps (BOUNDSCO)
Bock (1984): direct multiple shooting (MUSCOD)
Various error normsAlmost singular coefficient matrixImproved relaxation strategy
Improved robustness due to multipoint boundary value formulationReduced condition number by eliminating condensation
First-discretize-then-optimize code with multiple shooting
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Outline
• A short history on multiple shooting
• Multipoint-boundary-value-problem formulation
• A state constrained elliptic problem
• A state constrained parabolic PDE-ODE problem • A singular hyperbolic optimal control problem
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Abort Landing in a Wind Shear
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Maximal Minimum Altitude Optimal Solutionfor Different Wind Profiles
Montrone, P. 1991, Berkmann, P. 1995
max!
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Maximal Minimum Altitude Optimal Solution
bangsingular3rd order state constr
1st order state constr
first optimizethen discretize
byindirect
multiple shooting
control versus time: rate of angle of attack
altitude versus rangeof abort landing
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
A Very Complicated Switching Structure
3 bang-bang subarcs
2 singular subarcs
1 boundary subarc of a 1st order state constraint
1 boundary subarc of a 3rd order state constraint
1 touch point of a 3rd order state constraint
switching structure (7 pts, 12 add. var.):number of interiorboundary conditions:
4
2
6
plus 5 additional interior boundary conditions due to modellingplus 11 usual boundary conditions given or by optimality conditions
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Outline
• A short history on multiple shooting
• Multipoint-boundary-value-problem formulation
• A state constrained elliptic problem jointly with Michael Frey, Simon Bechmann & Armin Rund
• A state constrained parabolic PDE-ODE problem • A singular hyperbolic optimal control problem
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Model Problem: elliptic, distributed control, state constraint
Minimize
subject to
with
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Definition of active set and assumptions
Definition: active / inactive set / interface
Assumptionon addmissbleactive sets
No degeneracy.No active set
of zero measure.No common points
with boundary
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Reformulation of the state constraint
Transfering the Bryson-Denham-Dreyfus approach
Using the state equation
Optimal solution on given by data, but optimization variable
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Reformulation as set optimal control problem
Minimize
subject to
a posteriori check
inner outer
topology is assumed to be known
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Theorem:
For each admissible the objective is shape differentiable. The semi-derivative in the direction
is
Optimality system in the inner optimization of a bilevel problem
subject to the optimality system of the inner optimization problem
determines the interface
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
The Smiley example: rational initial guess
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
The Smiley example: bad initial guess
Algorithm can cope with topology changes to some extent
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany
Outline
• A short history on multiple shooting
• Multipoint-boundary-value-problem formulation
• A state constrained elliptic problem
• A state constrained parabolic PDE-ODE problem jointly with Armin Rund
• A singular hyperbolic optimal control problem
Multiple Shooting and Time Domain DecompositionMay. 6-8, 2013, IWR, Heidelberg, Germany