Modeling the output from a commercial chemical process using regression models from survival analysis

Abstract

This article is concerned with the modeling of the gas output of a commercial chemical plant using the coal sources as predictor variables. We consider the use of two models to incorporate these predictors; the Cox proportional hazards and accelerated failure time regression models. These models are chosen for their simplicity and for the ease with which the effects of explanatory variables can be interpreted. The contribution of this article lies therein that these models are used in the current context for the first time. We show, using both graphical and formal hypothesis testing procedures that these models (with a Weibull baseline distribution) fit observed gas production data well. We provide a discussion of the interpretation of the estimated model parameters and we comment on how these estimates can be of substantial practical value. The large scale of production from the chemical plant in question ensures that potential cost savings and increases in production associated with more accurate models are of great practical importance.