Reservoir Prediction by Machine Learning Methods on The Well Data and Seismic Attributes for Complex Coastal Conditions

The aim of this work was to predict the probability of the spread of rock formations with hydrocarbon-collecting properties in the studied coastal area using a stack of machine learning algorithms and data augmentation and modification methods. This research develops the direction of machine learning where training is conducted on well data and spatial attributes. Methods for overcoming the limitations of this direction are shown, two methods - augmentation and modification of the well data sample: Spindle and Revers-Calibration. The implementation and effectiveness of these methods are demonstrated on real data. Materials and methods. Data from 15 drilled wells, 930 seismic field attributes and 22 additional space attributes were used in this work. The proposed approach is based on binary classification algorithms with training on well data and includes the following sequence of applying methods: creating training datasets, selecting the features, creating a population of classification models, assessing the quality of the forecast, Reverse-Calibrating the seismic field position, creating additional datasets using Spindle method, assessing the contribution of the features in the forecast, combining models into an ensemble, the final forecast. Results. A prediction of the spatial development of reservoirs was made - a three-dimensional cube of calibrated probabilities of belonging of the studied space to the class of reservoirs was obtained. The estimation of the classification quality for the classification algorithms and changes of the forecast quality from application of the Reversal-Calibration and Spindle methods were done. Conclusion. Considering the difficulties in interpreting seismic data in coastal conditions, the proposed approach is a tool that is capable of working with the entire set of geophysical and geological data, extracting knowledge from a 159-dimensional space of spatial attributes, and making a forecast of the spread of facies with acceptable quality - the F1 measure for the collector class is 0.798 on average for the evaluation of drilling results under different geological conditions. It has been shown that the sequential application of the proposed augmentation methods in the implemented technology stack increases the quality of the collector forecast by more than 1.5 times relative to the original sample.