The current downturn in the industry has increased focus on production optimisation and cost efficiency. Organisations are now recognising opportunities to exploit and leverage data analytics to help deliver such improvements.
The paper describes how machine learning techniques have been successfully applied to improve production and processing performance.
It introduces a number of keys concepts relating to artificial intelligence and identifies the challenges of traditional data analysis techniques. The paper outlines a methodology in defining and evaluating suitable use cases for the application of machine learning and details both an upstream and downstream example of where the methodology has been successfully applied to optimise performance.
The first example presents how an operator transformed work practices to accurately predict valve actuator failures from anywhere between 12-18 months before the actual event using operational expertise, digitalisation and machine learning techniques.
Another example describes a natural gas transportation and processing system that transports gas from several offshore terminals to an onshore reception and processing terminal. Machine learning models were successfully developed to predict NGL Reid vapour pressure (RVP) at the processing terminal using historical (gas composition, process operating and RVP) data. RVP is a common measure of volatility as determined by the test method ASTM-D-323. The predictor model is valuable as it allows proactive export route management from the terminal, minimises potential interruption of production due to off-spec NGL and allows mitigation actions to be identified and implemented if high RVP is predicted to be likely.