Hybrid modelling with machine learning for overpressure and mud weight prediction

Unexpected high overpressures in the subsurface are still a significant challenge on the Norwegian Continental Shelf, even after decades of oil and gas exploration and production. Unpredicted overpressures may lead to challenges in drilling operations, such as unexpected gas into the well, loss of mud, and in the worst case, blow outs and loss of wells. Currently, drilling teams experience large uncertainties in prediction of pore pressure and wellbore stability, leading to extensive non-productive time and large costs. Today, several approaches are used to predict the pore pressures along a planned well path: either using data acquired during drilling, modelling approaches or combinations thereof. In practice, only a small amount of existing data, and often only from nearby wells, are considered, and geological models are not used in the predictions. The new idea is to use all data available, both historical and real time data, by using hybrid analytics and machine learning techniques to get better pressure prediction and thereby improved mud weight estimates. This novel approach aims at combining existing physical three-dimension geo-models and pressure models, with machine learning applied to all well data available in a sedimentary basin (on a hundred km scale). The project will move forward the state-of-art on pressure and mud weight window prediction. This will include machine learning using data from potentially hundreds of wells, to estimate key parameters that are used to predict well stability. The BigPressure project will be led by SINTEF Industry, in tight collaboration with researchers at NTNU, SINTEF Digital and University of Munich as well as industry partners Equinor and ConocoPhillips, who will contribute with industry perspectives on operation challenges and requirements. A PhD student at NTNU will be part of the project.

Unexpected high overpressures in the subsurface are still a challenge on the Norwegian Continental Shelf, even after decades of oil and gas exploration and production. Unpredicted overpressures may lead to challenges in drilling operations, such as mud loss, kicks, and in the worst case, loss of wells. Currently, drilling teams experience large uncertainties in prediction of pore pressure and wellbore stability, leading to extensive non-productive time (NPT) and large costs. Today, several approaches are used to predict the pore pressures along a planned well path: either using well logs, seismic, modelling approaches or combinations thereof. In practice, only a small amount of existing data, and often only from nearby wells, are taken into account, and the geological models are not used in the predictions. The idea is to use all data available, both historical and real time data, by using hybrid analytics and machine learning techniques to get better pressure prediction and thereby improved mud weight estimates. The novel approach aims at combining existing physical three-dimension geomodels and pressure models, with machine learning applied to all well data available in a sedimentary basin (on a hundred km scale). The project aims to move forward the state-of-art on pressure and mud weight window prediction, built on the existing tools, to include machine learning using data from potentially hundreds of wells, to estimate key parameters that are used to predict well stability. The BigPressure project will be led by SINTEF Industry, in tight collaboration with researchers at SINTEF Digital, with NTNU and University of Munich as well as industry partners Equinor and ConnocoPhillips, who will contribute with industry perspectives on operation challenges and requirements.