Issue link: https://industrialengineer.epubxp.com/i/905872
50 ISE Magazine | www.iise.org/ISEmagazine Inside IISE Journals research Using simulation to improve the quality of input-parameter estimation Models are often built to evaluate sys- tem performance measures or to make quantitative decisions. These models sometimes involve unknown input pa- rameters that need to be estimated sta- tistically using data. In these situations, a statistical method typically is used to estimate these input parameters, and the estimates are then plugged into the models to evaluate system output per- formances. However, the output performance estimators obtained from this ap- proach usually have large bias when the model is nonlinear and the sample size of the data is small. For instance, in a M/M/1/J queueing example, the aver- age queue length is an increasing con- vex function of the arrival rate. When the traffic intensity is high, an overes- timated arrival rate has a much larger impact on the estimated queue length than an equally underestimated one. This causes the queue-length estimator to have a large positive bias. As decisions typically are made ac- cording to the system performance measures, it is important to account for the bias of the performance estimators. This problem is investigated in the paper "A Simulation Based Estima- tion Method for Bias Reduction" by Jin Fang, a research associate from the Hong Kong University of Science and Technology, and Jeff Hong, a chair pro- fessor from City University of Hong Kong. They propose using simulation as a tool to learn the bias of the perfor- mance estimator and adjusting the input-parameter estimator based on the learned bias in situations where the models have no closed-form expres- sion and can be evaluated only through simulation. Several efficient algorithms are designed. Through the numerical examples, the authors find out that the biases of the performance estimators can be re- duced significantly when using the proposed simulation-based estimation (SBE) method. Compared with other bias reduction methods such as boot- strap and jackknife, the SBE method can reduce bias the most. This method can be implemented in a variety of systems, such as queueing systems and inventory systems, when nonlinear per- formance measures need to be evalu- ated from estimated input parameters. CONTACT: Jin Fang; jfang@connect.ust.hk; 86- 15618329659; Department of Industrial Engineer- ing and Logistics Management, the Hong Kong University of Science and Technology, Clear Water Bay, Kowloon, Hong Kong This month we highlight two articles from IISE Transactions. The first article studies how to address the uncertainties of input parameters and minimize the bias of outputs introduced by the input parameter estimation errors. To address this problem, the authors proposed using simulation to learn the bias of the performance estimator and adjusting the input- parameter estimator based on the learned bias. This is especially effective in situations where the models have no closed- form expression and can be evaluated only through simulation. The second article studies the multiple traveling salesmen problem to minimize the total distance traveled. The authors proposed a decomposition-based method, which outperforms existing methods and is easy to implement to solve real problems. These articles will appear in the January 2018 issue of IISE Transactions (Vol 50, No. 1). Jin Fang (left) and Jeff Hong studied the generic simulation problem of how to address uncertain input parameters and minimize output bias.