VFM (ESP ML-based) Package

This package estimates the liquid production rate of a well equipped with an ESP using a machine learning model

Usage

Run the workflows from this package to perform the calculations. The schedules may be adapted to use different contexts.

Predicting Liquid Production

The liquid production rate is calculated using a pre-trained ML model supplied with pump and well head data. If data is missing on a time step, the last complete data set is used. The predicted values are saved to the signal vfm liquid production rate.

To add the pre-trained ML Model to a workspace, download the linked .zip file. Then navigate to the Sources page in the PetroVisor Web App and create a new connection of type Excel. Press the Upload File button to choose and upload the model to the workspace.

Data Requirements

The following data is required on well level:
 variable speed drive current
 choke opening
 motor frequency
 pump intake pressure
 pump outlet pressure
 pump intake temperature
 well head temperature
 well head pressure

Retraining the Model

To train a PCA model with data from the workspace, run the workflow VFM (ESP ML): Train Model. This workflow trains a model and stores it to the workspace's file storage. The name of the newly trained model will be a concatenation of the Argument VfmLogModel and the name of the context used for running the workflow. If multiple contexts are used to run the workflow, a separate model will be created for each context. The workflow VFM (ESP ML): Predict will first look for a model for the passed in context. If none is found it will default back to using the global model.

Data Requirements

The following data is required on well level:
 variable speed drive current
 choke opening
 motor frequency
 pump intake pressure
 pump outlet pressure
 pump intake temperature
 well head temperature
 well head pressure
 liquid production rate