Skip to content

The Virtual FlowMeter - Enabling Lost Barrels to Be Tracked

A project presented by Philippe Solans, Sébastien Duchenne and Hugo Chavonnand

philippe_solans_exploration_production_total
Philippe Solans

Reservoir

A machine learning solution, this Virtual FlowMeter makes it possible to estimate the production of each oil well in real time without any intervention or additional sensors. This innovative solution will improve the operation and performance of our wells for the lowest possible cost.

The continuous virtual measurement of the flow rate of a well

Knowing the production of a well at all times is a basic requirement for managing its performance and monitoring lost barrels. The Virtual FlowMeter provides instant information on the flow rate of every well in operation. It estimates their production using machine learning algorithms fed with data from the well sensors and in-production tests.

This is a significant advance. The instantaneous flow rate is often unknown. Wells are not usually equipped with individual flowmeters: the equipment is expensive and takes up too much space on production platforms, which are generally overcrowded and are only fitted with a single flowmeter that is shared between multiple wells. The interval between two measurements or in-production tests on a single well can therefore vary from 15 days to several months.

Aerial view over Anguille platform, Gabon - Exploration & Production - Total

Machine learning boosting well performance

The Virtual FlowMeter will enable the real-time monitoring of each well to ensure it is producing at its full potential. This advance will benefit petroleum engineers by enabling them to operate wells more efficiently and ensure that planned targets are achieved. They will now be able to rapidly identify and correct unexpected events such as production deficits by adjusting the control parameters. The tool will also immediately measure the impact of an optimization action. The resulting decrease in the number of interventions needed will also help to reduce our environmental footprint.

The system is based on a machine learning algorithm. During the training phase, the solution collects measurements available on the actual flow rate of a well and connects to the well sensors to obtain “vital signs” in real time: pressure, temperature, etc. The algorithm learns a non-linear relationship between the vital signs and the actual flow rate. This is validated by correlating the calculated and measured flow rates to ensure that the predictive model is robust.

The Virtual FlowMeter is then ready to use this model to provide the virtual flow rate of the well continuously using data retrieved automatically from Total’s production databases.

Anguille Platform in Gabon - Exploration & Production - Total

A flow rate prediction that is over 90% reliable

The first prototype was developed in Python programming language in less than a month by a multidisciplinary team at Total Gabon combining well performance engineers, integrity engineers and geoscientists.

The prototype is currently undergoing pilot trials at Total Gabon. As 25% of production deficits are due to non-optimized behavior, this phase will make it possible to observe the savings achieved and adapt the tool to the operators’ requirements. The results from the Virtual FlowMeter are promising and we observe a confidence of 97% for liquid flow rates and 90% for oil flow rates.

The Virtual FlowMeter is scheduled to be rolled out at all Total Gabon wells by summer 2020. The project will also be considered by the Total Digital Factory for possible industrialization at Group level.

Deep Offshore

Technical Responses to New Challenges

Research & Development

The Driving Force Behind Total’s Competitiveness

Drilling & Wells

DrillX: Optimizing Drilling Using Digitization