A hybrid approach for accurate geothermal temperature prediction in the western region of Yemen

Abstract

The growing need for clean and reliable energy sources has propelled geothermal energy to the forefront of research. Its inherent advantages of baseload power generation and minimal environmental impact make it a highly attractive option. Nonetheless, effective utilization of geothermal resources relies heavily on the ability to accurately predict subsurface temperatures within geothermal reservoirs. In Yemen, a critical barrier to geothermal development is the lack of robust empirical correlations for predicting these crucial temperatures. This study presents a novel approach by developing a hybrid Particle Swarm Optimization Neural Network) PSONN (model to overcome this challenge. A dataset encompassing 1402 data points was collected from 108 geothermal wells located in Yemen's western region. The model incorporates key parameters influencing geothermal temperatures as inputs, including: Latitude; Longitude, Temperature gradient, Surface temperature, Depth, Elevation. Subsurface temperature serves as the model's output variable. The data were then strategically divided into two sets: 70 % designated for training the PSONN model and 30 % reserved for subsequent testing and validation of its predictive capabilities. The research successfully establishes the PSONN model as a highly effective tool for subsurface temperature prediction. This is supported by the exceptionally low Average Absolute Percent Relative Error (APRE) of 0.541, indicating a minimal deviation between predicted and actual values. Additionally, the low Standard Deviation (SD) of 0.11 signifies a high degree of consistency in the model's predictions. The findings suggest that the PSONN model achieves high predictive accuracy, with performance metrics (e.g., APRE = 0.541, SD = 0.11, R = 0.999) comparable to or exceeding those reported in prior studies (e.g., Haklidir, 2019; Altay et al., 2022). Further comparative analysis with existing methods is warranted to fully establish its relative advantages.