13.07.2015 Views

MPEC Problem Formulations in Chemical Engineering Applications

MPEC Problem Formulations in Chemical Engineering Applications

MPEC Problem Formulations in Chemical Engineering Applications

SHOW MORE
SHOW LESS

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

4.1 Automatic Reformulation of ComplementarityMPCC models written <strong>in</strong> the GAMS model<strong>in</strong>g language may be automaticallyreformulated as an NLP us<strong>in</strong>g several of the previously described complementaritysolution techniques. This can be accomplished with the use of the NLPECmeta-solver [8]. When called, NLPEC reformulates the complementarity constra<strong>in</strong>tsof a MPCC model with a user specified reformulation, <strong>in</strong>clud<strong>in</strong>g (17)-(20). NLPEC will then call a user-specified NLP solver to solve the reformulatedmodel. The f<strong>in</strong>al results from the NLP solver are then translated back <strong>in</strong>to theorig<strong>in</strong>al MPCC model. NLPEC verifies that the complementarities are satisfied<strong>in</strong> the result<strong>in</strong>g solution. This is particularly important when us<strong>in</strong>g the PF(ρ)formulation described above, as it may not converge to a stationary po<strong>in</strong>t of theMPCC if the penalty parameter is too small.4.2 Comparison of Solution Strategies on <strong>MPEC</strong>Lib<strong>MPEC</strong>Lib is a set of <strong>MPEC</strong> test problems ma<strong>in</strong>ta<strong>in</strong>ed by Dirkse [19]. The testset currently consists of 92 problems, which <strong>in</strong>clude both small scale modelsfrom the literature and several large <strong>in</strong>dustrial models.The performance and robustness of several NLP reformulations <strong>in</strong>clud<strong>in</strong>gPF(ρ) with ρ = 10, Reg(ǫ) with ǫ = 10 −8 , and NCP formulation x T y = 0,x, y ≥ 0 were compared on the <strong>MPEC</strong>Lib test set. CONOPT (version 3.14) andIPOPT (version 3.2.3) were used to the solve the result<strong>in</strong>g NLP problems. Allreformulations were performed automatically by the NLPEC solver describedearlier. Also <strong>in</strong>cluded <strong>in</strong> the comparison is the IPOPT-C solver, which is capableof process<strong>in</strong>g the complementarity constra<strong>in</strong>ts without reformulation. TheCONOPT and IPOPT runs were performed entirely with<strong>in</strong> the GAMS environment.However, IPOPT-C is not currently l<strong>in</strong>ked to GAMS. In order toperform the tests with IPOPT-C, the GAMS models were <strong>in</strong>stead convertedto the AMPL model<strong>in</strong>g language format. The problems were then run us<strong>in</strong>gAMPL and IPOPT-C.The results are presented <strong>in</strong> a Dolan-Morè plot. All results were obta<strong>in</strong>ed ona Intel Pentium 4, 1.8 GHz computer with 992 MB of RAM. This plot portraysboth robustness and relative performance of a set of solvers on a given problemset. Here we def<strong>in</strong>e a performance ratio calculated as:η p,s =t p,sm<strong>in</strong> {t p,s : 1 ≤ s ≤ n s }(23)where n s is the number of solvers, which are run over the n p problems <strong>in</strong> set P,and where t p,s is the solution time for solver s on problem p. For all solvers sthat fail to solve problem p, we set the performance ratio to an arbitrarily largenumber. The profiles are then def<strong>in</strong>ed by:p s (τ) = 1 n psize {p ∈ P : η p,s ≤ τ} (24)10

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!