This document outlines the configuration options to the PetroAI prediction pipeline that can be adjusted based on user input.  

Automated DCA 

PetroAI will auto-forecast all wells to get a best fir decline for oil, gas, and water streams. The following parameters can be changed to control the fits. 

  • Arps parameters that will set the range for all fluid streams 
  • Qi range (min and max) 
  • b-factor range (min and max) 
  • De range (min and max) 
  • Dmin range (min and max) 
  • Start Mode – default is to hang the curve from the peak month, however a specific month could also be selected 
  • Peak fluid – default option is to let the algorithm determine the peak fluid which sets the start date; however, this can be overridden to peak a specific fluid stream to define the start date 
  • GOR threshold - this option is used to determine the main fluid stream for each well 
  • Forecast duration – set the number of years to run the forecast, default is 40 years 

Interval Aliases 

Often the file name for a structure grid is not how that interval is referred to among the asset team. When using client-provided structure grids, PetroAI allows intervals to be aliased to align the data with the naming conventions used by a team  

Well Spacing and Well Interactions 

PetroAI performs well spacing calculations in stress space, interpolating the orientation of Shmax at each well (or well segment) location. This orientation creates the plane in which the offset wells are identified, and the horizontal distances calculated. Vertical distance is also calculated for all offset wells. Users can specify the criteria to find offset wells using five parameters: 

  • Maximum horizontal distance 
  • Maximum vertical distance 
  • Maximum direct distance 
  • Minimum lateral overlap – the amount of overlapping lateral required to pick up a well as an offset. 500 ft or 1000 ft is typically used.  
  • Segment count – define how many segments to divide the lateral when running the spacing, landing, and well interaction calculations. This number typically ranges between 1 and 5. 
  • Frac value (high and low) - this defines the gradient applied to the Frac Fingerprint. By default, it ranges from 0 to 1 to allow for co-stimulation between wells.   

Once offsets are identified they are classified as having a parent, child, or sibling relationship to the active well. These relationships are defined using an additional parameter on top of the offset criteria. 

  • Sibling days – the difference between the completion dates for two wells that applies the cutoff between two wells having a sibling relationship versus a parent – child relationship 

The impact of well spacing and well interactions can be captured in the calculated feature “Total Drainage” which is created using either a pre-trained Frac Fingerprint or through a simplified oval geometry.  

  • Use Oval – can be true or false  
    • When false, a trained Frac Fingerprint and vertical stress profile are required 
    • When true, the frac width and height can be specified 

Model Features 

Users can select which features to try in a model. These can include some combination of engineering (lateral length)computed (total drainage), and subsurface features (PHIE)Multiple model versions can be created to test different features. 

Users can also specify different filters to constrain the training data. For example, changing the cutoff on completion year can exclude older wells that might not be relevant for the designs being consideredAny column in the well header or production data can be used to filter the training data.  

Predictions and Sensitivities 

PDP Predictions

In addition to making time series predictions on PDP wells, sensitivities can be run on each well in the data set. 

  • Predicted well performance over a range of proppant intensities 
  • Predicted well performance over a range of fluid intensities 
  • Predicted well performance over a range of lateral lengths 

PUD Predictions 

Sensitivities are run across the entire extent of the geomodel in 10-square mile cells to produce expected well performance for a variety of DSU configurations.  

  • Predicted parent well performance over a range of proppant intensities 
  • Predicted parent well performance over a range of fluid intensities 
  • Predicted parent well performance over a range of lateral lengths 
  • Predicted well performance and diagnostics for a co-developed DSU at various well spacings (e.g. 4, 6, 8 WPS) 
  • Predicted well performance for a co-developed DSU with set wine-rack configurations 

Inventory Predictions 

Similar to PDP predictions, inventory locations can be loaded into PetroAI and time series predictions made. These predictions account for any well interactions that might exist.