Predicting the frequency of abnormal events in chemical process with Bayesian theory and vine copula

Abstract Chemical accidents, such as an explosion, are of low frequency and high consequence (e.g. casualties, significant economic losses, pollution). Due to the shortage of accident data, recently, precursor data have received much attention in chemical risk analysis. Usually, in chemical processes, an abnormal event can be seen as a precursor, which can propagate into near-miss, incident or even accident. The abnormal event frequency (AEF) is defined as the number of abnormal events in a time interval, which can be an early indicator of risk. In this paper, an AEF predicting model based on Bayesian theory and D-vine copula is proposed. Generally, a chemical process is managed in shifts by several teams. The AEFs vary with different experience and operational skills of the operator teams. Furthermore, the previous operating team has an effect on the following operator teams and the effects are asymmetric between two teams, hence, D-vine copula is employed to describe the dependence with much flexibility. Finally, the proposed method is applied to a case study of 4-group-3-shift, and the simulation result shows that it has a better performance compared to conventional approaches.