A multi-objective parameter calibration approach

Belter, D.S. (2013) A multi-objective parameter calibration approach.

Abstract:This master thesis presents a diagnostic study on scalarized, multi-objective automatic calibration of the bag-gage batching process of a Vanderlande system at Schiphol South Terminal. Several simulation and emulation steps are necessary within the design and implementation of such a system. The problem we study is how to incorporate feedback from later implementation stages into initial simulation or emulation models that are based on a high abstraction level. Our main objective is to automate the calibration methodology for large-scale simulation models that have to be matched with realistic values from follow-up project stages or the implemented on-site system itself. This automatic calibration can help to reduce overall project lead-time and enhance the validity of simulations and emulations. We test the performance of several calibration methodol-ogies according to their convergence speed and accuracy of parameter estimation in comparison to an initial parameter estimation based on histogram frequency matching. The test calibration methods are Random Search, Latin Hypercube Sampling, a Simulated Annealing adaptation, and a combination approach that merg-es Simulated Annealing with a Latin Hypercube (LHS-SA). All test approaches performed significantly better than the initial, manual parameter estimation. The singular use of our Simulated Annealing modification shows the best convergence characteristics of all tested calibration algorithms regarding both speed and accuracy. However, its cooling scheme is vital for a successful calibration attempt. It was not possible to prove that syn-ergy effects of the LHS-SA method exist, but still we suspect them to be present. This might be due to the fact that the researched problem structure did not include many local minima. However, we cannot clearly verify this by our observations. Our Simulated Annealing adaptation has shown to be an effective automatic meth-odology for the calibration of multi-objective problems which include several tunable parameters.
Item Type:Essay (Master)
Faculty:BMS: Behavioural, Management and Social Sciences
Subject:85 business administration, organizational science
Programme:Industrial Engineering and Management MSc (60029)
Link to this item:http://purl.utwente.nl/essays/66022
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page