Machine-learning in oil and gas exploration: A new approach to geological risk assessment

Abstract

During risk assessment, exploration geoscientists routinely evaluate various individual risk segments to estimate the chance of success of a given prospect. Usually, this estimation is based on subjectivity, unconscious biases that may lead to inconsistent evaluations. In this research project, the goal is to constrain these biases using cognitive advisors, supported by machine learning algorithms, to overcome the frequent subjectivity and have a fair prospect ranking. This new methodology is based on a new metric - the Level of Knowledge (LoK) - that relies on the available data and geological model to characterize a given prospect. This parameter is anchored to a powerful source of information - the Knowledge Base - a database that structures peer-reviewed inputs and other unstructured information, such as research papers or reports, giving more accurate advices to geoscientists. Overall results show a positive response by testers, which identify the value of having a normalized metric on available knowledge and experience (LoK), as the base for having consistent and fair assessments on their Probability of Success evaluations. As the system is continuously trained, oil and gas companies that integrate this system would benefit of having a consistent portfolio management, thus, supporting exploration strategies.