登录查看更多内容

How can you use machine learning to predict permeability in well logs?

由人工智能和领英社区提供技术支持

Permeability is a key parameter for reservoir characterization and production optimization in the oil and gas industry. However, measuring permeability directly from well logs is challenging, as it depends on complex pore-scale interactions and fluid properties. Machine learning (ML) offers a promising alternative to traditional empirical or analytical models, as it can learn from data and capture complex nonlinear patterns. In this article, you will learn how to use ML to predict permeability in well logs, using a simple workflow and some common tools.

此文章中的业界达人

由社区从 2 条内容中精选。了解更多

Ese Omatsone

Professional Engineer, Business/Data Analyst, Energy Researcher (Renewables), Storyboard Creator
Chamberlain Mbah

Senior Data Scientist | AI, Machine learning Consultant | Python Developer| Empowering Businesses with AI & Machine…

1 Data preparation

The first step in any ML project is to prepare your data. You need to have a dataset of well logs with corresponding permeability measurements, either from core analysis or well tests. You also need to perform some quality control, such as removing outliers, filling missing values, and normalizing the data. You can use tools like Pandas, NumPy, or Scikit-learn in Python to manipulate and process your data. You should also split your data into training, validation, and test sets, to avoid overfitting and evaluate your model performance.

添加您的观点

Chamberlain Mbah

Senior Data Scientist | AI, Machine learning Consultant | Python Developer| Empowering Businesses with AI & Machine Learning Expertise | Crafting Solutions in Python
举报内容
Like in any other domain, you need a trained model. Training requires features and a target. You train a model with existing well log data where permeability is known, correlating it with attributes like porosity, grain size, and fluid saturation. Once trained, the model can infer permeability in new logs, helping to assess reservoir quality and optimize extraction strategies. As simple as that. The real challenge is in the amount of information the logs carry. Machine learning models will only learn patterns in your data and hence are bounded by how reach the logs are in terms of predictive information about permeability.

已翻译

赞

2 Feature selection

The next step is to select the most relevant features for your ML model. Features are the variables that influence the target variable, which in this case is permeability. You can use domain knowledge, correlation analysis, or feature importance methods to identify the best features for your problem. For example, you can use the mutual information score from Scikit-learn to rank the features according to their dependency with permeability. You should also consider the availability and reliability of the features in different wells and scenarios.

添加您的观点

3 Model selection

The third step is to choose the best ML model for your problem. There are many types of ML models, such as linear regression, decision trees, neural networks, or support vector machines. Each model has its own advantages and disadvantages, depending on the complexity, interpretability, and scalability of the problem. You can use tools like Scikit-learn, TensorFlow, or PyTorch in Python to implement and train different ML models. You should also compare the models based on metrics like mean squared error, R-squared, or coefficient of variation.

添加您的观点

4 Model evaluation

The fourth step is to evaluate your ML model on the test set, which is the data that you have not used for training or validation. This will give you an unbiased estimate of how well your model can generalize to new data. You can use tools like Matplotlib, Seaborn, or Plotly in Python to visualize the results and analyze the errors. You should also check the assumptions and limitations of your model, such as linearity, homoscedasticity, or independence. You can use tools like Statsmodels or Scipy in Python to perform statistical tests and diagnostics.

添加您的观点

5 Model deployment

The final step is to deploy your ML model into production, where it can be used to predict permeability in new wells or scenarios. You need to have a robust and reliable infrastructure to store, manage, and update your data and model. You also need to have a user-friendly interface to interact with your model and get the outputs. You can use tools like Flask, Django, or Streamlit in Python to create web applications or dashboards for your model. You should also monitor and maintain your model, to ensure its accuracy and relevance.

添加您的观点

6 Here’s what else to consider

This is a space to share examples, stories, or insights that don’t fit into any of the previous sections. What else would you like to add?

添加您的观点

Ese Omatsone

Professional Engineer, Business/Data Analyst, Energy Researcher (Renewables), Storyboard Creator
举报内容
This is actually the first thing to consider as soon as we start the data gathering, data cleansing and feature selection processes. Permeability is a scale- and fluid-dependent rock property. It has the dimensions of L-squared, like a measure of (cross-sectional) area. Production or injection rates and pressures are influenced not just by a measurement of permeability, but by the product of permeability and pay thickness. This product, also known as flow capacity, has the dimensions of L-cubed, like a unit volume. This rock property is highly scale-dependent (core scale, near wellbore- scale or reservoir drainage area-scale). It is crucial to understand this when using data from different sources in selecting and training ML models.

已翻译

赞

Oil & Gas

+ 关注

给文章评分

我们借助人工智能创建了此文章。您认为这篇文章怎么样？

很棒不太好

举报此文章

查看全部

How can you use machine learning to predict permeability in well logs?

1

2

3

4

5

6

1 Data preparation

2 Feature selection

3 Model selection

4 Model evaluation

5 Model deployment

6 Here’s what else to consider

Oil & Gas

给文章评分

感谢您的反馈

更多Oil & Gas相关文章

更多相关阅读内容