Intelligent deepwater energy development: Flow assurance monitoring and smart decision-making system

Deepwater hydrocarbon resources (oil, gas, and hydrate) constitute an important part of the world's existing and emerging energy landscape. The high hydrostatic pressure, low temperature, and complex geological structures of deepwater environments present significant challenges for energy development. The development of flow assurance monitoring and intelligent decision-making systems to quantify the risk of hydrate blockage is a prerequisite for safe and efficient development. In this work, the hydrate growth and blockage under varying liquid loading, pump speed, pipe inclination, subcooling, and oil-water ratio were investigated using a high-pressure visual flow loop. A database containing 27 sets of tests with about 510,000 pieces of data was created. The evolution of differential pressure in oil-gas-water-hydrate multiphase flow was found to have two stages. Stage I, differential pressure remains stable or rises slightly with increasing hydrate concentration. Stage II, differential pressure changes abruptly after the hydrate reaches the critical concentration. Therefore, the prediction of hydrate blockage was defined as an unsupervised anomaly detection problem. The Transformer-LSTM-VAE (TLV) blockage prediction model was developed based on the Variational Auto-Encoder framework. The Transformer and LSTM models were utilized as encoder and decoder, respectively. Two uneven weight adjustment methods were proposed to achieve earlier warnings of blockage. In systematic comparisons with three classic and two state-of-the-art machine learning models, the TLV model consistently performed the best with an F1-score of 0.969. The TLV model demonstrated the potential of deep learning methods in deepwater energy development.