Since 2015, one of Total’s ambitions has been to promote a strong data culture among its collaborators to move towards a data-driven and a digitalized industry. In that context, the Refining and Chemical branch has built a data analytics center to deal with industrial data science matters.
Using new data science methodology, tools and infrastructure, machine learning models have been implemented to provide the Business with real-time estimations of various parameters for our industrial assets. In this presentation we will focus on the estimation of the percentage of gasoil remaining in the residue of a Crude Distillation Unit (CDU). Based on conditional parameters (flow, temperature, pressure …), a data-driven model has been built in R and Python and implemented in the refinery’s PI System. In order to anticipate possible model degradation over time, an online retraining has also been developed.
In this presentation, the main focus will be to present the technical frame to develop such a workflow: needed infrastructure, data science methodology and algorithms, useful tools and software.
In order to get the best out of this presentation, attendees are expected to be familiar with the general framework of a data-driven project without necessarily being an expert in data science.