/GeothermalDatathon

The U.S Department of Energy is developing Enhanced Geothermal Systems as a solution for renewable energies. The Utah Forge Project is currently searching for the optimum well placement of a production well that enables the maximum possible net energy for 20 years. ‘GeotherML’ team participated in this challenge, analyzed data provided by SPE - PIVOT, and delivered a solution that includes the use of Deep Learning. The development of this initiative covers Exploratory Data Analysis with feature engineering, data modeling and evaluation metrics.

Primary LanguageJupyter Notebook

PIVOT 2022 Geothermal Datathon: SPE - Gulf Coast Section

- Team:

GeotherML
Presentation Link: click here

- Contents:

  • Exploratory Data Analysis

    • Overview of dataset
    • Check for missing data
    • check for columns without meaningful information
    • Determine inputs and outputs based on project scope
    • Check for correlation between features
    • Overview of embedded Time-series columns
    • Feature engineering: Average quarterly time-series to yearly time-series and filter to 20year lifespan
    • Import Second Dataset and concatenate with initial dataset
    • Export all dataframes for future Data Modeling

  • Data Modeling: Power Enthalpy

    • Import inputs and outputs
    • Train / Test split
    • Scale Data
    • Create Neural Network
    • Fit the Model

  • Data Modeling: Power Output

    • Import inputs and outputs
    • Train / Test split
    • Scale Data
    • Create Neural Network
    • Fit the Model

  • Data Modeling: Extracted Thermal Power

    • Import inputs and outputs
    • Train / Test split
    • Scale Data
    • Create Neural Network
    • Fit the Model

  • Evaluation Metrics

    • MAE
    • MSE
    • RMSE
    • Explained variance score
    • Compare mean values per year: Predictions Vs Real