Machine Learning Toolkit
Machine Learning Toolkit on i2G offers the most advanced predictive algorithms with a fully controllable parameter set-up in an intuitive graphic user interface. The toolkit supports multi-zone and multi-well selection to optimize petrophysical tasks with:
Robust algorithms designed by geoscientists, machine learning experts
Reproducible results by allowing seed number input
Flexibility in configuring model inputs and architectures
Easy-to-use and adjustable workflow interface
Availability of different model validation metrics
Classification Model
The built model can used to predict discrete properties such as lithofacies, depofacies, flow units or rock types. The module offers diversified ways to assess the model: loss and accuracy crossplots as well as confusion matrix. Similar to regression model, user has an ability to perform data filtering and input ranking.
Here is the list of supporting classification methods:
•
Decision Tree Classification
•
KNN Classification
•
Logistic Classification
•
Neural Network Classification
•
Random Forest Classification
Non-linear regression
This is a nonlinear approach to model the relationship between the input curves and the target curve. The relationships are modeled using nonlinear predictor functions, so called nonlinear models, whose unknown model parameters are estimated from the data. The nonlinear relationship function is defined by the user, where the variables x1, x2, … correspond to the input curves.
Regression Model
The built model can used to predict quantitative petrophysical/geochemical/geomechanical properties as well as generate synthetic output for missing data or bad-hole intervals
Multi perceptron algorithm enables building multi-layer neural network with non-linear activation functions to solve complex geological settings. Here is the list of supporting regression methods:
•
Decision Tree Regression
•
Huber Regression
•
Lasso Regression
•
Linear Regression
•
Neural Network Regression
•
Random Forest Regression
•
SVM Regression
•
Xgboost Regression
Self-organizing Map (SOM)
Self-organizing Map is a unique tool for facies classification by both supervised and unsupervised modes. The tool provides a robust workflow from model construction to model validation which is critical to make sure the output brings actual geological meaning. Emsemble Boosting SOM and Distributed Ensemble SOM algorithms are also two supported algorithms which provides more options to deal with a wide range of geological complexities.