Predicting real-time geothermal well flow rate and enthalpy with machine learning techniques

Abstract

Geothermal energy is a sustainable energy source offering reliable and renewable energy solutions. However, accurately measuring geothermal well output like flow rate and enthalpy for wells that produce a two-phase fluid remains challenging due to the complexity and infrequency of traditional methods. This thesis addresses these issues by continuing the work of developing a real-time method to measure flow rate and enthalpy from geothermal wells without interrupting operations. The focus is on accurately estimating geothermal fluids' flow rate and enthalpy using advanced rule-based models and machine learning techniques.This research integrates data-driven approaches for continuous monitoring and early detection of well performance changes by using measurements from Landsvirkjun's geothermal operations conducted in 2019, 2020, 2021, and 2023. The study employs a specialized differential pressure orifice plate meter setup at Theistareykir and Bjarnarflag Geothermal Power Plants, providing detailed measurements critical for the models.The most effective model employed Density-Based Spatial Clustering of Applications with Noise (DBSCAN) for noise reduction, Recursive Feature Elimination with Cross-Validation (RFECV) for precise feature selection, and Random Forest Regression (RFR) with five key features, achieving a Root Mean Square Error (RMSE) of 0.011. This approach can significantly enhance the efficiency and accuracy of geothermal power production measurements, offering insights into real-time monitoring and operational optimization