Non-linear modeling of a production process by hybrid Bayesian Networks

摘要

This paper shows how non-linear functions can be approximated by hybrid Bayesian networks. The basic idea is to make a piecewise linear approximation with several base points. This approach is applied to an engineering domain and the accuracy is compared to Gibbs sampling. Great accuracy is shown even at non-continuous functions. Due to the general underlying principle, it is possible to adapt this type of network to other domains.