Machine learning based truncation point estimation in steady-state simulation

Date

2023-9-08

Author

Girit, Burak

Metadata

Show full item record

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.

Item Usage Stats

212
views

532
downloads

Cite This

In this study, we focus on estimating the truncation point for solving the initialization bias problem encountered in output analysis for steady-state simulations by using machine learning methods. Since the initial conditions of simulation do not represent the steady-state, biased data originating from the initial state must be eliminated in order to analyze the simulation output properly. A truncation point in simulation output data has to be determined for eliminating this initialization bias. In this study, the truncation point estimation capabilities of multilayer perceptron regressor, long short-term memory, and conditional recurrent neural networks are investigated. In order to train these three neural networks and test their performances, the second order autoregressive model and M/M/1 queueing system model are used to generate data representative of the simulation output. Moreover, these three machine learning methods are compared with the conventional truncation estimation methods MSER and MSER-5. Experimental results show that the multilayer perceptron regressor network has superior performance compared to other methods, in terms of truncation point estimation error and coverage of the confidence intervals for steady-state expected values. However, the long short-term memory and conditional recurrent neural networks cannot learn effective truncation point estimation with the network configurations used.

Subject Keywords

Initialization bias in steady-state simulation, Truncation point estimation, Multilayer perceptron regressor, Long short-term memory, Conditional recurrent neural network, Machine learning

URI

https://hdl.handle.net/11511/105374

Collections

Graduate School of Natural and Applied Sciences, Thesis