Institute for Systems Research Technical Reports
Permanent URI for this collectionhttp://hdl.handle.net/1903/4376
This archive contains a collection of reports generated by the faculty and students of the Institute for Systems Research (ISR), a permanent, interdisciplinary research unit in the A. James Clark School of Engineering at the University of Maryland. ISR-based projects are conducted through partnerships with industry and government, bringing together faculty and students from multiple academic departments and colleges across the university.
Browse
Search Results
Item Randomized Difference Two-Timescale Simultaneous Perturbation Stochastic Approximation Algorithms for Simulation Optimization of Hidden Markov Models(2000) Bhatnagar, Shalabh; Fu, Michael C.; Marcus, Steven I.; Bhatnagar, Shashank; Marcus, Steven I.; Fu, Michael C.; ISRWe proposetwo finite difference two-timescale simultaneous perturbationstochastic approximation (SPSA)algorithmsfor simulation optimization ofhidden Markov models. Stability and convergence of both thealgorithms is proved.Numericalexperiments on a queueing model with high-dimensional parameter vectorsdemonstrate orders of magnitude faster convergence using thesealgorithms over related $(N+1)$-Simulation finite difference analoguesand another two-simulation finite difference algorithm that updates incycles.
Item Comparing Gradient Estimation Methods Applied to Stochastic Manufacturing Systems(2000) Mellacheruvu, Praveen V.; Fu, Michael C.; Herrmann, Jeffrey W.; ISRThis paper compares two gradient estimation methods that can be usedfor estimating the sensitivities of output metrics with respectto the input parameters of a stochastic manufacturing system.A brief description of the methods used currently is followedby a description of the two methods: the finite difference methodand the simultaneous perturbation method. While the finitedifference method has been in use for a long time, simultaneousperturbation is a relatively new method which has beenapplied with stochastic approximation for optimizationwith good results. The methods described are used to analyzea stochastic manufacturing system and estimate gradients.The results are compared to the gradients calculated fromanalytical queueing system models.These gradient methods are of significant use in complex manufacturingsystems like semiconductor manufacturing systems where we havea large number of input parameters which affect the average total cycle time.These gradient estimation methods can estimate the impact thatthese input parameters have and identify theparameters that have the maximum impact on system performance.
Item Optimal Multilevel Feedback Policies for ABR Flow Control using Two Timescale SPSA(1999) Bhatnagar, Shalabh; Fu, Michael C.; Marcus, Steven I.; ISROptimal multilevel control policies for rate based flow control in available bit rate (ABR) service in asynchronous transfer mode (ATM) networks are obtained in the presence of information and propagation delays, using a numerically efficient two timescale simultaneous perturbation stochastic approximation (SPSA) algorithm. Numerical experiments demonstrate fast convergence even in the presence of significant delays and a large number of parametrized parameter levels.Item Gradient Estimation of Two-Stage Continuous Transfer Lines Subject to Operation-Dependent Failures(1998) Fu, Michael C.; Xie, Xiaolan; ISRThis paper addresses the gradient estimation of transfer linescomprising two machines separated by a buffer of finite capacity. A continuous flow model is considered, where machines are subject tooperation-dependent failures, i.e., a machine cannot fail when it is idle. Both repair times and failure times may be general, i.e., they need not be exponentially distributed.The system is hybrid in the sense that it hasboth continuous dynamics, as a result of continuous material flow, and discrete events: failures and repairs. The purpose of this paper is to estimate the gradient of the throughput rate with respect to the buffer capacity. Both IPA estimators and SPA estimators are derived. Simulation results show that IPA estimators do not work, contradicting the common belief that IPA always works for continuous flow models.Item Application of Perturbation Analysis to the Design and Analysis of Control Charts(1997) Fu, Michael C.; Hu, Jian-Qiang; ISRThe design of control charts in statistical quality control addresses the optimal selection of the design parameters such as the sampling frequency and the control limits; and includes sensitivity analysis with respect to system parameters such as the various process parameters and the economic costs of sampling. The advent of more complicated control chart schemes has necessitated the use of Monte Carlo simulation in the design process, particularly in the evaluation of performance measures such as average run length. In this paper, we apply perturbation analysis to derive gradient estimators that can be used in gradient-based optimization algorithms and in sensitivity analysis when Monte Carlo simulation is employed. We illustrate the technique on a simple Shewhart control chart and on a more complicated control chart that includes the exponentially- weighted moving average control chart as a special case.