Loading...

Development of Non-deterministic Methods in Metamodel-based Simulation Optimization

Moghaddam, Samira | 2016

871 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 49121 (01)
  4. University: Sharif University of Technology
  5. Department: Industrial Engineering
  6. Advisor(s): Mahlooji, Hashem; Eshghi, Kourosh
  7. Abstract:
  8. In recent years, simulation optimization methods have been developed to solve complicated problems that cannot be solved by mathematical programming methods. In simulation optimization methods, first the problem is modeled by simulation tools and then by applying optimization tools the optimal combination of input variables that optimizes the simulation output is determined. Although simulation optimization has attracted researchers’ attention in recent years, most of the works presented do not consider uncertainty in simulation models. This becomes our motivation in this study to develop uncertain methods in metamodel-based simulation optimization based on minimax methods that are applicable to constraint problems.
    First, we develop an algorithm based on minimizing the expected value of the response for semi-expensive problems (problems with simulation time 1 to 5 minute). The algorithm has both the model-based and the metamodel-based characteristics. In addition, a new optimization algorithm based on the particle swarm optimization metaheuristic is developed in which strategies are embedded for moving toward the optimal solution and improving the metamodel simultaneously.
    Moreover, we present a new robust approach by using the Bertsimas and Sim methodology to find the robust counterpart for the objective function obtained by Kriging metamodel. We find the robust counterpart problem for the three most common correlation functions: triangular, exponential, and Gaussian. Our method is then extended to the constrained problems to address robustness in feasibility of the model.
    As robust methods give conservative solutions, we present two new robust simulation optimization methods using ϕ-divergence function in which the probability of uncertain parameters is considered to reduce the conservatism of solutions.
    To show the performance of the proposed methods, the problem of train timetabling in metro is simulated. Tehran metro, line 1 is studied as a case and the results are presented
  9. Keywords:
  10. Simulation Optimization ; Kriging Metamodel ; Particles Swarm Optimization (PSO) ; Robust Optimization ; Kriging Metamodel ; Bertsimas Approach ; Train Scheduling ; Metamodel-Based Algorithm ; Semi-Expensive Simulation

 Digital Object List

 Bookmark

No TOC