Missed Net Pay Zones Mature Oilfieds Via Injection Of Expert Knowledge in Deep Learning Algorithms

In this work we pursued the goal of an automate cognitive system development capable for searching missed net pay zones within wells to extend a brownfield lifecycle via the involvement of new reserves into the field development. Additional research was dedicated to studying the possibility of knowledge transfer across different fields and the construction of the ranking model allowing fast expertise conduction of proposed intervals and evaluation of the proposed method on mature assesses. The proposed approach is based on deep learning and artificial neural networks architectures trained in a supervised mode using a provided human well logs interpretation. Our approach also utilizes transfer learning procedures in order to reuse knowledge extracted from the oilfield with sufficient data and improve the predictive qualities of the model on a target oilfield. Additionally, we proposed a ranking model that simulates expert decision-making process and evaluates oil saturation potential of proposed intervals by sorting it by a confidence level. Developed method was evaluated at the one of Gazpromneft brownfields, located in Western Siberia, Yamalo-Nenets region. The model was trained on a data from this field and its analogues with subsequent reinterpretation of the whole well log volume. Several hundreds of new net pay intervals were proposed and post-processed by ranking model. Then the list of proposed intervals was analyzed by an expert group including number of geologists, petrophysicists and reservoir engineers. Significant part of these intervals after detailed and comprehensive evaluation were marked as missed during previous manual well log interpretation conducted by petrophysicist and taken for the following fieldwork. Produced results confirmed applicability of proposed algorithm and proved its capability for localization of previously unrecognized net pay intervals.