13.07.2015 Views

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

Assessment and Future Directions of Nonlinear Model Predictive ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

462 P. Kühl et al.4.1 Min-Max NMPC Simulation ResultsFor the numerical solution <strong>of</strong> (5) the direct multiple shooting approach was used.The NMPC settings are the same as described in Section 3.2. The confidencefactor γ was slowly increased from zero to the desired level. For each optimizationwith a respective γ, the previous result has been used to initialize the states <strong>and</strong>control. In the case presented here, the initial amount <strong>of</strong> catalyst K 1 is assumedto be uncertain. The st<strong>and</strong>ard deviation <strong>of</strong> K 1 is 0.17 g. The confidence levelhas been chosen to be 99.7 %, i.e. we have γ = 3 to obtain the 3σ-interval.Figure 2 shows the simulation results for the robust version <strong>of</strong> the NMPC for thenominal amount <strong>of</strong> catalyst <strong>and</strong> an increased amount. The solution is comparedto the solution <strong>of</strong> the nominal NMPC for the nominal catalyst amount chargedto the reactor. One can see that the robust solution strongly resembles thenominal solution. Only, the dosing is stopped earlier. This ensures that lessA is accumulating in the reactor <strong>and</strong> leaves a safety margin to the adiabatictemperature as can be seen in the lower right graph. When the robust NMPCcontroller is confronted with a plant-model mismatch it reacts by dosing A morecarefully. The singular sub-arc becomes flatter than computed with the nominalNMPC. Eventually, less A is present in the reactor <strong>and</strong> the temperature peakgets lower. This also leads to a slower consumption <strong>of</strong> B which, however, isaccounted for by a higher reactor temperature at the end <strong>of</strong> the batch so thatthe final productivity losses are very small.The safety margins for the reactor temperature <strong>and</strong> the adiabatic temperatureare the main feature <strong>of</strong> the robust NMPC scheme. The fact that they are also[mol/s][K]43210330320310300x 10 −3Dosing rate u(t)0 1000 2000 3000time [s]Reactor temperatureSafetymargin2900 1000 2000 3000time [s][mol][K]76543212−butanol00 1000 2000 3000time [s]360350340330320310300NMPC: 5.01 gRNMPC: 5.01 gRNMPC: 5.10 gAdiabatic temperatureSafetymargin2900 1000 2000 3000time [s]Fig. 2. Simulation results <strong>of</strong> robust (RNMPC) compared with nominal NMPC. TheRNMPC is shown for the nominal <strong>and</strong> a perturbed initial amount <strong>of</strong> catalyst K 1.Dashed lines denote constraints. Note the RNMPC safety margins.

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

Saved successfully!

Ooh no, something went wrong!