Multiple input and multiple output simulation metamodeling using Bayesian networks

Abstract

This paper proposes a novel approach to multiple input and multiple output (MIMO) simulation metamodeling using Bayesian networks (BNs). A BN is a probabilistic model that represents the joint probability distribution of a set of random variables and enables the efficient calculation of their marginal and conditional distributions. A BN metamodel gives a non-parametric description for the joint probability distribution of random variables representing simulation inputs and outputs by combining MIMO data provided by stochastic simulation with available background knowledge about the system under consideration. The BN metamodel allows various what-if analyses that are used for studying the marginal probability distributions of the outputs, the input uncertainty, the dependence between the inputs and the outputs, and the dependence between the outputs as well as for inverse reasoning. The construction and utilization of BN metamodels in simulation studies are illustrated with an example involving a queueing model.

Machine learning-based metamodels for sawing simulation

“…This additional information proved to be useful at improving the accuracy of the prediction on the KPI. We find, among other recently used metamodeling techniques, Bayesian networks (Poropudas et al 2011) and least-square regression (Salemi et al 2012).…”

Machine learning-based models of sawmills for better wood allocation planning

“…Regression analysis, Kriging interpolation, radial basis functions and neural networks are among the usual used metamodeling techniques (Barton and Meckesheimer, 2006;Kleijnen, 2015). We find, among other recently used metamodeling techniques, genetic programming (Can and Heavey, 2012), Bayesian networks (Poropudas et al, 2011), and least square regression (Salemi et al, 2012). These methods share common ground with statistical (machine) learning techniques.…”