Machine learning models for forecasting planned levels of mineral production

Authors: Dergachev A. O., Borovskoy I. G., Kruchinin V. V.

Annotation: The oil and gas industry of the Russian Federation, a powerful driver of the country's economic development, directly depends on the speed of commissioning new mining fields. The decline in the accuracy of hydrocarbon production forecasts by Russian companies is a consequence of the deteriorating quality of their resource base. This study evaluates the effectiveness of ma-chine-learning-based predictive models for forecasting hydro-carbon production volumes. A method for training predictive neural network models is described, incorporating a dataset of geological, geophysical and design indicators for the develop-ment of oil and gas fields. Missing geological and geophysical data were reconstructed using various augmentation methods.

Keywords: data normalization, production facility, hydrocarbon feedstock, data augmentation, cnn1d, mlp, lstm, gru