A data-driven online truncation method for transient bias reduction in steady-state simulations. (September 2019)

• Record Type: Journal Article
• Authors: Zhang, Nan; Yuan, Jun; Ng, Szu Hui
• Is Part Of: Computers & industrial engineering. Volume 135(2019)
• Page Start: 723
• Page End: 745
• Publication Date: 2019-09
• Subjects: Simulation -- Steady state -- Transient bias -- Online truncation methods; Engineering -- Data processing -- Periodicals; Industrial engineering -- Periodicals; 620.00285
• DOI: 10.1016/j.cie.2019.06.034
• Languages: English
• ISSNs: 0360-8352

Abstract

Highlights: Develop a data driven online truncation method (DOT) to identify and cut off transient period for steady-state simulations. DOT can provide the real-time truncation position when the simulation model is still running. DOT does not require human interactions in truncation process. DOT is efficient and accurate to detect the truncation position. Abstract: In steady-state simulations, the conclusions of a steady-state performance are biased if transient bias remains in the simulation outputs. The transient bias is caused by an unrealistic initial condition when starting a simulation. This paper discusses one of the bias mitigation methods that reduces transient bias by deleting the initial biased observations' phase, namely, truncation methods. The majority of existing truncation methods usually cut off the biased phrase after simulation outputs are collected. This wastes the computer budget and time. In the literature, there are few studies trying to explore the truncation approaches that can provide the real-time truncation position in an online simulation process, where the biased phrase is truncated when the simulation model is running. Such online truncation methods determine the truncation position in real time, saving the computer budget and consequently improving the simulation efficiency. However, the existing online truncation methods usually lack automation and have poor performance since they usually underestimate truncation positions. This paper proposes a new online truncation method that enables the data-driven approach to help automate the truncation process and utilizes the hypothesis test to determine the truncation position. This new truncation approach named Data-driven Online Truncation method (DOT) is able to automate transient bias truncation for any simulation analysts to increase the reliability and accuracy of the results. The validity and efficiency of the proposed online approach are tested and verified with artificial data sets, M/M/1 queuing systems and serial production lines and are further compared with other existing online truncation methods. The results of the numerical tests show that this proposed algorithm is an effective and robust online truncation method.