Machine learning model for predicting accidents at gas production facilities

Abstract: Emergency situations lead to disruptions of workflows, regard-less of how quickly these emergencies are resolved, therefore the ability to predict such situations would be extremely useful in many fields. The paper implements an approach to prepro-cessing data from a SCADA database using averaging and correlation analysis. A machine learning (ML) model is then trained on the preprocessed data to classify the current state of the node as an emergency precursor to simulate a real work-flow. The model achieved high accuracy score and other metrics without fine tuning and hyperparameter optimization, thus con-firming the possibility of using ML models for the task.

Keywords: classifier, technological process, fault detection, scada, data preprocessing, machine learning