Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Soft Sensor for Bottom-Hole Pressure Estimation in Petroleum Wells Using Long Short-Term Memory and Transfer Learning

M. A. Fernandes, E. Gildin, M. A. Sampaio

TL;DR

This work tackles estimating flowing bottom-hole pressure (BHP) in offshore wells when Permanent Downhole Gauges are unavailable, by developing a data-driven soft sensor that uses surface and topside measurements. It compares Long Short-Term Memory (LSTM) against Multilayer Perceptron (MLP) and Ridge regression, and introduces transfer learning to adapt models across fields with differing characteristics. Using a 13-year, multi-field offshore dataset from Brazil's pre-salt basin, the study demonstrates generally sub-2% mean absolute percentage error, with LSTM delivering the strongest performance, especially under transient operating conditions, and transfer learning enabling effective cross-field deployment. The approach offers a cost-effective, real-time alternative to physical sensors and supports applications in digital twins, anomaly detection, and intelligent well management in diverse reservoir and flow scenarios.

Abstract

Monitoring bottom-hole variables in petroleum wells is essential for production optimization, safety, and emissions reduction. Permanent Downhole Gauges (PDGs) provide real-time pressure data but face reliability and cost issues. We propose a machine learning-based soft sensor to estimate flowing Bottom-Hole Pressure (BHP) using wellhead and topside measurements. A Long Short-Term Memory (LSTM) model is introduced and compared with Multi-Layer Perceptron (MLP) and Ridge Regression. We also pioneer Transfer Learning for adapting models across operational environments. Tested on real offshore datasets from Brazil's Pre-salt basin, the methodology achieved Mean Absolute Percentage Error (MAPE) consistently below 2\%, outperforming benchmarks. This work offers a cost-effective, accurate alternative to physical sensors, with broad applicability across diverse reservoir and flow conditions.

Soft Sensor for Bottom-Hole Pressure Estimation in Petroleum Wells Using Long Short-Term Memory and Transfer Learning

TL;DR

This work tackles estimating flowing bottom-hole pressure (BHP) in offshore wells when Permanent Downhole Gauges are unavailable, by developing a data-driven soft sensor that uses surface and topside measurements. It compares Long Short-Term Memory (LSTM) against Multilayer Perceptron (MLP) and Ridge regression, and introduces transfer learning to adapt models across fields with differing characteristics. Using a 13-year, multi-field offshore dataset from Brazil's pre-salt basin, the study demonstrates generally sub-2% mean absolute percentage error, with LSTM delivering the strongest performance, especially under transient operating conditions, and transfer learning enabling effective cross-field deployment. The approach offers a cost-effective, real-time alternative to physical sensors and supports applications in digital twins, anomaly detection, and intelligent well management in diverse reservoir and flow scenarios.

Abstract

Monitoring bottom-hole variables in petroleum wells is essential for production optimization, safety, and emissions reduction. Permanent Downhole Gauges (PDGs) provide real-time pressure data but face reliability and cost issues. We propose a machine learning-based soft sensor to estimate flowing Bottom-Hole Pressure (BHP) using wellhead and topside measurements. A Long Short-Term Memory (LSTM) model is introduced and compared with Multi-Layer Perceptron (MLP) and Ridge Regression. We also pioneer Transfer Learning for adapting models across operational environments. Tested on real offshore datasets from Brazil's Pre-salt basin, the methodology achieved Mean Absolute Percentage Error (MAPE) consistently below 2\%, outperforming benchmarks. This work offers a cost-effective, accurate alternative to physical sensors, with broad applicability across diverse reservoir and flow conditions.
Paper Structure (16 sections, 10 equations, 18 figures, 10 tables)

This paper contains 16 sections, 10 equations, 18 figures, 10 tables.

Table of Contents

  1. Introduction
  2. Contextualization and Objectives
  3. Problem Statement
  4. Traditional Methods
  5. Current Research
  6. Objectives and Contributions
  7. Methodology
  8. Database
  9. Data Conditioning and Feature Engineering
  10. Aspects of the Machine Learning Methods
  11. Evaluation Metrics
  12. Results
  13. Hyperparameter Tuning and Model Selection
  14. Model Testing for Field 1
  15. Evaluating Transfer Learning for Field 2
  16. ...and 1 more sections

Figures (18)

  • Figure 1: Schematic of an offshore production system, highlighting the positioning of the most relevant equipment and sensors (reproduced from b3).
  • Figure 2: Schematic representation of a gas lift valve operating within an oil well. Note: image enhanced using ChatGPT.
  • Figure 3: Flowchart depicting the proposed methodology, including base model training and transfer learning.
  • Figure 4: Boxplots showing the probability distributions of the variables for the two fields.
  • Figure 5: Dataset partitioning.
  • ...and 13 more figures