Installing increased numbers of sensors on existing and new well stock does not necessarily result in increased production. New methods of analysis must also be developed to capitalize on the new data streams to maximize safety and value delivery. A novel approach to modelling gas coning, which has been difficult to characterize using first-principles models, has been developed for one of BP’s assets. It is based on of the integration of several data-driven models representing different aspects of a well’s performance characteristics. These models use the well data that is captured by existing sensors.
In this presentation we describes how the data-driven approach has been developed and successfully tested on a North Sea reservoir operated by BP. The suite of data-driven models provides the capability to predict the fluid and gas rates for use in short-time-loop optimizations.