Machine Learning and Data Science for Upstream Professionals

Introduction

Objectives

By the end of this training course, participants will learn to:

• Core concepts of machine learning and data science
• Identifying existing bottlenecks for machine learning methods applied in your professional domain
• Choosing the most appropriate machine learning methods to solve a particular problem
• Applying the main machine learning methods in practice

Training Methodology

The participants of this training course will receive thorough training on the subjects covered by the course outline with the instructor utilizing a variety of proven adult learning teaching and facilitation techniques. This will include presentations, live feedback and interactions, quizzes, and group discussion.. The online course is conducted online using MS-Teams/ClickMeeting.

Who Should Attend?

A reservoir engineer, geologist, or Petrophysicist, and keen to obtain a fundamental understanding and practical knowledge on scientific programming, data science, and machine learning.

Course Outline

Day 1:

• Introduction to Machine Learning ecosystem
• Python crash course
• Data wrangling (using Pandas and SQL)
• Data visualisation

Exercises:

• Production data analysis and visualisation
• Data preparation for material balance calculations
• Reservoir simulation model QC
• Well log data visualisation

You will learn how to:

• Confidently use Python programming language and the main machine learning libraries to solve different problems from upstream domain
• Create a powerful and reusable workflow for production data analysis from different sources (local files and production databases) that can be applied for small and large oil and gas fields
• Quickly prepare production and pressure data for material balance calculation for the reservoirs of high-level of complexity (multiple compartments and pressure datums) in the format of industry-standard software (PETEX MBAL)
• Analyse a large number of reservoir simulation runs in an efficient way, quickly getting insights into history matching quality and forecasting results
• Easily create high-quality visualisation of different kinds of field and well data (production, pressure, well log) to simplify the data analysis and get ready-to-use plots for presentations and reports

Day 2:

• Numerical optimisation
• Statistics refresher
• Exploratory data analysis
• Uncertainty evaluation and decision making

Exercises

• Decline curve analysis
• PVT data preparation for reservoir simulation
• Volume-in-place probabilistic estimation
• Static model upscaling
• Waterflood optimisation

You will learn how to

• Apply different numerical optimisation methods to solve practical problems from the reservoir engineering domain (fitting rate-time data to understand the reservoir depletion mechanism, matching the reservoir pressure gradient with PVT data for consistent reservoir simulation model initialisation)
• Perform smart upscaling of the fine grid static model into the coarse grid reservoir simulation model with precise control of the upscaling process and finding a trade-off between model dimensionality reduction and the level of geological details preservation
• Perform the probabilistic volume-in-place estimation taking into account the uncertainty of input parameters to quickly evaluate volumetrics without building a full-scale geological model
• Allocate water and gas injection volume between injection wells to maximise oil production using the optimal number of reservoir simulation runs

Day 3:

• Machine learning introduction
• Dimensionality reduction methods
• Clustering methods
• Anomaly detection methods

Exercises

• Electrofacies identification based on well log data
• Static model realisations screening
• Numerical well testing

You will learn how to

• Confidently apply machine learning terminology and identify technical and business requirements for successful application of machine learning methods
• Choose the most suitable machine learning method to solve a particular problem from the upstream domain depending on the type of the problem, data availability, data quality and solution requirements
• Perform screening of static model scenarios to simplify the history matching process, reduce the number of simulation runs and efficiently evaluate the impact of geological uncertainty on the production forecast
• Identify the optimal number of electrofacies for a modelling study to guide the distribution of properties in the reservoir model
• Prepare the pressure data for pressure transient analysis (PTA) by automatically removing error pressure measurements to reduce the amount of manual efforts and build a fully automatic workflow for PTA

Day 4:

• Machine learning core concepts
• Regression methods
• Tuning of machine learning models

Exercises

• Production forecast of unconventional reservoir
• Saturation pressure prediction

You will learn how to

• Design and perform machine learning study to ensure the solution quality and reproducibility of the modelling results
• Apply on practice and understand the main concepts of machine learning modelling: train/test split, cross-validation, objective function definition, bias-variance trade-off, hyperparameters tuning
• Predict the performance of a new well and optimise the well completion design for unconventional reservoirs without building a sound physics-based reservoir simulation model
• Develop a powerful data-driven model incorporating available fluid studies and predict the saturation pressure with high accuracy for the reservoirs with missing key PVT experiments
• Automatically find the combination of machine learning model parameters to simplify the model tuning and reduce the amount of manual efforts

Day 5:

• Classification methods
• Neural networks and deep learning
• Advanced machine learning topics:
  • Imbalanced datasets
  • Interpretability of machine learning models

Exercises

• Lithofacies identification
• Screening of enhanced oil recovery (EOR) methods

You will learn how to

• Explain machine learning modelling results to technical and business audience to perform QA/QC solution and support decision making
• Develop a robust classification model for lithofacies identification based on well logs for wells without core data
• Create an enhanced oil recovery screening model that allows incorporating different sources of information (PVT, SCAL, geological data), performing screening of a company's fields portfolio in an efficient way and identifying the most suitable EOR method for a particular field