Numerical investigations on a geothermal reservoir using fully coupled thermo-hydro-geomechanics with integrated RSM-machine learning and ARIMA models

In this research work, the thermo-hydro-geomechanical (THM) model is upgraded by including dynamic fluid, rock and fracture properties, and integrated machine learning (ML)-response surface model (RSM)-autoregressive integrated moving average (ARIMA) model to enhance the heat production from a geothermal reservoir. Changes in reservoir pressure, temperature, stress, strain, and rock physical, mechanical and thermal properties were examined effectively. Maximum pressure, low-temperature zone found in the neighborhood of injection well. Thermal stress and strain were operating highly near the injection well, and mechanical stress and strain were effective in the rest of the area. Mathematical equations utilized for the dynamic variations in the rock physical, mechanical and thermal properties were keener toward the low-temperature region during the heat extraction. An integrated ML-RSM-ARIMA model was successfully integrated into numerical simulations and implemented. In ML models, the deepNN-3 model with 15 hidden layers, each with 15 neurons, showed better predictions. It is used to predict production temperature for the desired values of input parameters. ARIMA (2,1,1) model was used to forecast the production temperature obtained from the deepNN-3 model with ±95% confidence interval. Therefore, the developed numerical simulations approach with integrated ML-RSM-ARIMA can help examine geothermal production with greater accuracy.