Thermo-economic optimization of an enhanced geothermal system (EGS) based on machine learning and differential evolution algorithms

Hot dry rock (HDR) is considered as a promising low-carbon alternative to fossil fuels, but the remaining eco-nomic challenges are leading to its unsuccessful exploitation. Therefore, incorporating economic indicators into consideration is essential to optimize an enhanced geothermal system (EGS). However, the conventional opti-mization approaches based on numerical simulations are time-consuming and a global optimal operation strategy is hard to determine. In this study, an optimization framework based on Artificial Neural Network (ANN) and Differential Evolution (DE) is proposed with considering a levelized cost of electricity (LCOE) being an economic performance indicator to optimize a three-horizontal-well EGS in the Qiabuqia field. Specifically, four different ANN models are constructed to predict different geothermal productivities to substitute numerical models. Based on these ANN models, a DE optimization process is conducted to determine an optimal LCOE under two field operating constraints, followed by a performance comparison between the resulting optimal geothermal system and 2,150 randomly created cases using a numerical simulator. The results show that these ANN models all achieve a coefficient of determination R2 higher than 0.996, demonstrating their predictive abilities and potential as surrogate models. The determined optimal parameters configuration brings a promising LCOE of 0.0376 $/kWh which is around 50 % of a local electricity cost, and this is the lowest LCOE among all random cases. Importantly, the proposed framework can significantly save operation time by 36,000 times compared with the numerical simulation method. The proposed method provides a valuable reference for the geothermal system studied, and it can also be effectively applied to other energy systems, thereby facilitating their optimal development.