Increasing the Effectiveness of Simulation Models
| Autor: | Wenchi Chiu, 邱文琪 |
|---|---|
| Rok vydání: | 2007 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 95 When undertaking the estimation of a performance measure, we may interested in three types of problems: (I) how should one obtain a good estimator of the performance measure and the estimator’s quality? (II) what are the relationships between inputs and outputs? (III) what can we do if the occurrence rate is too low during the experiments? In this dissertation, we focus on increasing the quality of estimation. The first part is related to the estimation of variance of sample mean of simulation models: providing the quality measure of the estimated values more precisely. The second part is related to the estimation of metamodel: providing a better metamodel to know the inherent relationship between input and output variables. The third part is related to estimation of the bit error rate: using importance sampling to overcome the low occurrence rate during the experiment. Specifically, we propose a combined estimator which can more precisely estimate the variance of sample mean, a random number assigning method which can estimate a more precise metamodel, and the mixed basing distribution for estimating bit error rate using importance sampling technique. For the first part, estimating the variance of the sample mean is a prototype problem in steady-state simulation. We propose a bias-aware mechanism which attempts to minimize the variance of an estimator subject to a bias constraint – a goal that differs from that of minimizing mse (sum of variance and bias squared), in which case there would be no explicit bias constraint. Specifically, we use linear combinations of estimators based on different batch sizes to satisfy the bias constraint; and then we minimize variance by choosing appropriate linear combination weights. We illustrate the use of this mechanism by presenting bias-aware linear combinations of several variance estimators, including non-overlapping batch means, overlapping batch means, and standardized time series weighted area estimators. We also evaluate our mechanism with Monte Carlo examples. For the second part, estimating the simulation metamodel, which is a functional relationship between the mean response of the simulation model and a set of simulation inputs, is an advanced simulation problem. We propose a five-class variance swapping rule, which classifies all variances of the effect estimators into five classes, for linear metamodels of 2k factorial designs. The proposed rule is a generalization of all existing variance swapping rules (VSRs) and is a better VSR than the existing ones in that it makes a finer distinction among all effects, provided that the most important effects have possible minimal variance and all the lower-interaction effect estimators have smaller variances than that for the highest-interaction effect. For the third part, we are interested in estimating the BER for signal transmission in digital communication systems. Since BERs tend to be extremely small, it is difficult to obtain precise estimators based on the use of crude Monte Carlo simulation techniques. In this research, we review, expand upon, and evaluate a number of importance sampling variance reduction techniques for estimating the BER. We find that a mixture of certain “tailed”distributions with a uniform distribution produce estimators that are at least competitive with those in the literature. Our comparisons are based on analytical calculations and lay the groundwork for the evaluation of more-general mixture distributions. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|