Machine Learning Predictions (MLP)

Machine Learning is based on the idea that systems can learn from data, identify patterns, and make decisions with minimal intervention. A key workflow for this feature is the Production Prediction Workflow which reveals optimal engineering parameters to feed into the model. The Production Prediction Workflow document details the steps for creating a production prediction model using the Zones table, and also shows how to set optimal engineering constants and grid data to run the model on the Grids table. For an overview of running the model on a different table, see

Machine learning algorithms can be leveraged to predict production performance, create sweet spot maps, and estimate optimum engineering parameters for your next wells. Computations and templates are created automatically, and the independent variables are selected. The model also provides easy-to-understand error metrics and visualizations that help you determine the best parameters for the model.

In this application of machine learning, the general process is the following:

  1. Data table: Determine what data table you want to use to build the analytical model. Then select the input columns from the selected data table.
  2. Limit data using markings: Use only the marked rows in the table for calculations. The default is to use all rows in the table.
  3. Attribute to calculate: Identify the output (the column you want to calculate). This column likely has sparse data. The machine learning algorithm will calculate predicted values using your input data.
  4. Input attributes: Select the attributes from the selected data table that you want to include in the making of the model. Input attributes should be numeric, not text, although text attributes are accepted.
  5. Operation type: Regression or Classification. The operation type depends on the attribute to calculate. If the attribute is numeric, then the operation type is regression. If the attribute is text, then the operation type is classification. Integer values can be either regression or classification. It is common to assign an attibute to an integer value. For example, for facies, shale=1, sandstone=2, carbonates=3. Each operation type has its own set of visualizations.
  6. Run Multicollinearity Analysis (optional). This process helps determine the best inputs for the machine learning model, or conversely, the attributes to exclude from the model for optimal results

  7. Specify thePrediction Algorithms & Settings to use. Advanced settings are also available to adjust for those who have knowledge and experience in the machine learning field. Please note that the default values have been extensively tested and generally yield the best results.

  8. Select the output options.
  9. Click Run.

The model will run using the default parameters and settings.

The machine learning algorithm automates the model building process, running through multiple iterations until results are optimized. After the predicted data has been calculated, multiple visualizations are created for you to analyze the results. The predicted values are written to the data table.

After examining the mean average error (MAE) plot and the Attribute Analysis chart, you can adjust the input attribute selection and possibly the hyperparameters, and run the model again. Iterate on this process until the results are optimized, although the relative gain will likely decrease with each iteration.

Your saved model can be applied to other data tables. See Run Machine Learning Model for more information on applying the model to other tables.

The remaining steps are optional. However, the output of these options provide interesting results and can help you understand the process and calculations: