Тип публикации: статья из журнала
Год издания: 2025
Идентификатор DOI: 10.32014/2025.2518-170X.479
Ключевые слова: oil and gas industry, pumping unit, forecasting, remaining equipment life, machine learning, recurrent neural network, regularization
Аннотация: Relevance. The oil industry is one of the key areas of the mining industry. It has a significant impact on the economy of the state, providing various branches of industrial production with oil refining products. To ensure the normal functioning of enterprises it is important to organise continuous transportation of hydrocarbons usПоказать полностьюing pumping units. This approach allows to determine the residual life of the pump unit and to carry out its timely maintenance before failure. Objective. The aim of the work is to ensure uninterrupted transportation by means of predictive maintenance based on big data processing technology and machine learning methods. Methods. Recurrent neural networks LSTM and GRU were used as mathematical models to determine the residual life of the pump unit in this paper. Results and Conclusions. The effectiveness of the modernised recurrent neural networks was demonstrated by comparing them with traditional machine learning methods (PCR and Random Forest) on different data variations. The comparative analysis demonstrated the significant performance of recurrent neural network based models, especially the LSTM model. An improvement in prediction accuracy was shown in comparison with PCR and Random Forest. The average percentage improvement in the four metrics was 36.35 % and 25.21 % using a smaller sample (n = 10000), and 31.86 % and 25.64 % using a larger sample (n = 20000)
Журнал: Известия Национальной академии наук Республики Казахстан. Серия геологии и технических наук
Выпуск журнала: Т.1, №469
Номера страниц: 107-127
ISSN журнала: 22245278
Место издания: Алматы
Издатель: Национальная академия наук Республики Казахстан