Assessment of Data Driven Techniques for Flow Rate Estimation in Sub Sea Oil Production

EasyChair Preprint 14652

Abstract

Accurate measurement of flow rate of the multiphase flow of oil, gas and water from the oil wells, is an important part of the oil and gas industry. This enables the safe operation and proper optimization of the production. With the increasing availability of process data, machine learning algorithms are used to create models for various applications. The application of these algorithms for flow rate estimation provides a more accurate representation of the oil and gas production process. In this paper, two oil wells and ten machine learning algorithms are evaluated. Long short-term memory (LSTM) provides the best results with Mean absolute percentage error of 1.96% for Well 1 and 1.56% for Well 2. In addition, the effects of noise on the models are explored. Median filter with window size of three provides good noise reduction. The uncertainty of the predictions are quantified using 95% confidence intervals in XGBoost model.

Keyphrases: Data-driven estimations, Machine Learning Techniques, measurement noise, uncertainty quantification

@booklet{EasyChair:14652,
  author    = {Neville D'Souza and Carlos Pfeiffer and Gaurav Mirlekar},
  title     = {Assessment of Data Driven Techniques for Flow Rate Estimation in Sub Sea Oil Production},
  howpublished = {EasyChair Preprint 14652},
  year      = {EasyChair, 2024}}