In order to solve the problem that the simulation-based optimization method usually requires a large number of simulations, in this paper, after comparing the prediction performance of different machine learning methods, a surrogate model based on MLP was developed to reduce the computational cost, which was then combined with genetic algorithm to develop an optimization method of development parameters for geothermal doublets in heterogeneous geothermal reservoirs. Through the case study of a doublet system, the reasonability and efficiency of the developed optimization method of development parameters were demonstrated. The results show that when given a certain position of production well, the surrogate model-based optimization method of development parameters can accurately find the optimal placement of injection well, the rate of production and injection, and the temperature of recharge water with lower computational cost.

geothermal doublet system  /  machine learning  /  optimization of development parameters  /  surrogate models  /  genetic algorithm