Alternative prediction models for data scarce environment

Abstract

Effective accident prediction is needed in the chemical process industries to facilitate risk management during plant operations. This is however hampered by the unavailability of data needed for accident modelling purposes, and models that are based on distribution theory are used as they require the least amount of data. This article discusses the application of grey modelling approach and its combination with Bayesian network. The models are applied to two case studies, i.e. a process vessel and an LNG facility. The results obtained are compared to that of Poisson model. Results show that the hybrid first-order grey model with Bayesian network BG(1,1) is most accurate, followed by the grey models G(1,1) and G(2,1), with the Poisson model trailing behind. The results illustrated the potentials of grey modelling approach in dealing with scarce data conditions