A Retrospective Framework for Enhancing Explainability and Reliability of Predictive Models in Digitalized Oil&Gas Projects

The given Retrospective Analysis Framework delivers a structured methodology for getting trustworthy predictive models in Oil and Gas Industry. The combination of process domain knowledge with the analytics of mode behavior, we achieve a framework that ensures that the model’s digital and physical aspects of performance are under continuous control. The framework encompasses four modules – input-output relationship analysis, root cause analysis, drift detection, and error analysis – each part responsible for a different part of model health and data integrity checks.

The framework has successfully been applied to an oil refinery model which was behaving abnormally, and has helped tremendously to identify the root cause of the issue affecting model performance. The approach transforms predictive modelling into a continuous, adaptive process where the model can be monitored constantly, and any issue can be resolved effectively and in a short time to ensure consistent operation.