Geothermal potential of a petroleum mature field using machine learning methods: Ecuadorian Amazon region

Producing oil and extracting geothermal energy from the same reservoir is one type of the multigeneration systems that are garnering global attention. The aim of this research is to use machine learning methodologies to quantify the potential geothermal energy yield within a mature petroleum reservoir located in the Amazon region of Ecuador. The paper covers the complete geothermal simulation process of an area in the central-east region of Sacha Field. The study compares the performance of permeability calculation using the Kozeny and Carman algorithm in conjunction with Principal Component Analysis against permeability is calculated using the K-Nearest Neighbor density estimation algorithm. Historical matching is achieved using the Particle Swarm Optimization Algorithm, emphasizing a water production objective function. Additionally, the potential quantification compares the Volumetric method with Monte Carlo for the simulation of three scenarios until the year 2040. The results demonstrate excellent history matching using the Kozeny–Carman equation with Principal Component Analysis methodology. The historical production of water until the year 2009 was 2.625 million [std m3]1 and the simulation result was 2.757 million [std m3]. The percentage of error in the volume is 5.02% with the historical production approach. A small area in the central east region of the reservoir has the potential to generate a maximum power of 6.5 MWt2 based on the Volumetric methodology, which aligns with the average produced power of 0.66 MWt, depending on which of the three future scenarios up to 2040 is applied. •Combined well logs with core samples to estimate permeability using upscaling, Kriging, and Kozeny-Carman equations, with enhanced results from Principal Component Analysis.•Compared permeability predictions using the K-nearest neighbor density estimation algorithm KNN with those from the Kozeny-Carman equations.•Utilized the Particle Swarm Optimization Algorithm (PSO) to compare two reservoir models—one using Kozeny-Carman permeabilities and the other using K-nearest neighbor density estimation (KNN)—to determine which model better predicts reservoir performance.•The model using Kozeny-Carman permeabilities outperformed the model using the K-nearest neighbor density estimation algorithm.