Aplicação de modelos deep learning na estimação de vazão de petróleo em poços offshore com sistema de bombeio centrífugo submerso

Abstract

The fluid flow measurement is a fundamental activity for the oil and gas industry. The correct produced volumes mensuration provides a good reservoirs management, reducing production losses, guiding plans of the production system optimization and production flow methods. The use of flow estimation techniques in real time using Virtual Flow Metering (VFM) has shown to be a promising field due to the provided results precision and their low-cost implementation. Deep learning models have been applied successful in the oil and gas industry. Combining technological advances and the great importance of fluid measurement, the study aims to develop a model for the flowrate of liquid applying an approach combined of Long Short-Term Memory (LSTM) models and hydrodynamical modelling. The data used were power, frequency, pressure and were collected from two offshore wells with electric submersible pumps (ESP) in Northeast region of Brazil. The LSTM results compare favorably with the results of hydrodynamical modeling and increases its powerful, they can be useful joint to accurately estimate the flowrate behavior in real time in transient and steady states and to forecast the flowrate for a sequence of future time instants, supporting better production management. It is expected that the results obtained with the LSTM neural networks can be integrated with other technologies of Industry 4.0 and contribute to the digital transformation of the oil and gas industry.