On analysing time series of accidents and incidents at hazardous production facilities

Safe operation of any complex distributed systems is largely defined by the quality of the event analysis and prediction tools. Aside from the external hazards, the so-called hazardous production facilities (HPFs) pose significant threats. Accidents at such facilities are a subject of constant analysis and concern of operating organisations. At the same time, the accident statistics accumulated over the period of operation of such facilities are often heterogeneous. Accidents and incidents at HPFs occur at different times, against different forecast backgrounds, which complicates the construction and verification of digital models of such facilities. This paper proposes an algorithm for preprocessing time series of observations to extract data that can be subsequently used to build and train predictive models with the required accuracy. The proposed approach can be implemented by means of the R language, that in many respects has become a standard for statistical calculations.