WSC 2005 Final Abstracts

Enhancing Evolutionary Algorithms with Statistical Selection Procedures for Simulation Optimization
Axel Thümmler and Peter Buchholz (University of Dortmund, Department of Computer Science)

Abstract:
In this paper, we present an evolution strategy for the optimization of simulation models. Our approach incorporates statistical selection procedures that efficiently select the best individual, where best is defined by the maximum or minimum expected simulation response. We use statistical procedures for the survivor selection during the evolutionary process and for selecting the best individual from a set of candidate best individuals, a so-called elite population, at the end of the evolutionary process. Furthermore, we propose a heuristic selection procedure that reduces a random-size subset, containing the best individual, to at most a predefined size. By means of a stochastic sphere function and a simulation model of a production line, we show that this procedure performs better in terms of number of model evaluations and solution quality than other state-of-the-art statistical selection procedures.

A New Optimization Heuristic for Continuous and Integer Decisions with Constraints in Simulation
Mufit Ozden (Miami University)

Abstract:
In this paper, a new metaheuristic optimization approach is developed for the mixed integer decisions with constraints within a simulation model. Each decision variable is handled by an optimizer that uses a machine learning technique. At the beginning of each iteration, the decisions are selected randomly from their decision distributions. The performance evaluation is estimated during a short simulation run. The optimizers modify their selection-distributions for the decisions that prove to be “good” performance judged against an advancing threshold value. Then, a new set of decisions is generated for the next run. When the average performance reaches a good competency, the threshold value is advanced to a higher level. Thus, the optimizers are forced to learn toward the optimal solution. In this paper, after brief explanation of the approach, we present an application to a challenging engineering problem dealing with pressure-vessel design.

Simulation-based Optimization for Repairable Systems Using Particle Swarm Algorithm
Talal M. Al-Khamis and Mohamed A. Ahmed (Kuwait Univeristy)

Abstract:
We describe an approach based on particle swarm optimization (PSO) for determining the optimal allocation of spares as well as repair resources while satisfying a desired availability constraint. The proposed method expands the original PSO algorithm to handle stochastic constraints and discrete decision variables. Computational results show that the proposed approach is efficient for determining the optimal choice of spares and repair channels for multi-echelon repairable-item inventory systems.

Approximate/Perfect Samplers for Closed Jackson Networks
Shuji Kijima and Tomomi Matsui (University of Tokyo)

Abstract:
In this paper, we propose two samplers for the product-form solution of basic queueing networks, closed Jackson networks with multiple servers. Our approach is sampling via Markov chain, but it is NOT a simulation of behavior of customers in queueing networks. We propose two of new ergodic Markov chains both of which have a unique stationary distribution that is the product form solution of closed Jackson networks. One of them is for approximate sampling, and we show it mixes rapidly. To our knowledge, this is the first approximate polynomial-time sampler for closed Jackson networks with multiple servers. The other is for perfect sampling based on monotone CFTP (coupling from the past) algorithm proposed by Propp and Wilson, and we show the monotonicity of the chain.

Nonparametric Estimation of the Stationary M/G/1 Workload Distribution Function
Martin B. Hansen (Aalborg University)

Abstract:
In this paper it is demonstrated how a nonparametric estimator of the stationary workload distribution function of the M/G/1-queue can be obtained by systematic sampling the workload process. Weak convergence results and bootstrap methods for empirical distribution functions for stationary associated sequences are used to derive asymptotic results and bootstrap methods for inference about the workload distribution function. The potential of the method is illustrated by a simulation study of the M/D/1 model.

An Efficient Performance Extrapolation for Queuing Models in Transient Analysis
Mohamed A. Ahmed and Talal M. Al-Khamis (Kuwait University)

Abstract:
In designing, analyzing and operating real-life complex systems, we are interested, however, not only in performance evaluation but in sensitivity analysis and optimization as well. Since most systems of practical interest are too complex to allow the analytical solution of totally realistic models, these systems must be studied by means of Monte Carlo simulation. One problem with Monte Carlo analysis is its expensive use of computer time. To address this problem, we propose an efficient technique for estimating the expected performance of a stochastic system for various values of the parameters from a single simulation of the nominal system. This technique is based on the likelihood ratio performance extrapolation (LRPE). We provide numerical experiments that demonstrate how the proposed technique significantly outperform the likelihood ratio performance extrapolation technique in the context of the Markovian queueing models in transient analysis.