关键词: 浅层地热能')">

浅层地热能, 同轴套管式换热器, 数值模拟, 机器学习, 随机森林模型

Abstract:

In order to achieve the efficient development and utilization of shallow geothermal energy resource and solve the environmental problems such as haze caused by traditional heating relying on fossil energy. It also solves the problems of unclear operation parameters of heat exchangers and the optimal matching method of ground formation parameters in practical engineering, and consumes manpower and material resources. In this paper, a combination of numerical simulation and machine learning algorithm is adopted. Firstly, a heat transfer model and numerical model of shallow coaxial casing geothermal system were established based on COMSOL, and the effects of different physical parameters, operating parameters and length of geothermal well on heat extraction performance were explored. Then, the random forest (RF) model was used to predict the geothermal system mining temperature of different parameter sets based on a machine learning algorithm, and the importance of the corresponding characteristic influencing factors was obtained in the RF model, which further verified the accuracy  of RF in predicting the geothermal mining temperature. The results show that the injection flow rate and injection temperature of heat exchange fluid have a great influence on the temperature of geothermal production. The injection flow rate is maintained between 0.5 and 1.0 m3/h, and the temperature difference between the formation and injection temperature can improve the heat recovery efficiency of the system. The random forest model has high accuracy in geothermal temperature prediction, the coefficient of determination （R2） is close to 0.99, and the root mean square error RMSE is only 0.17. The feature importance diagram further verifies that different injection temperature and injection velocity have signifcant impact on the shallow coaxial casing geothermal system.

Key words: shallow geothermal energy, coaxial tube heat exchanger, numerical simulation, machine learning, random forest model