Time-lapse Geophysical Monitoring And Deep Learning Prediction Of Hot Dry Rock Reservoir

With the fossil energy crisis,environmental pollution and the gradual implementation of China's double carbon plan,hot dry rock(HDR),an essential new renewable clean energy,will play a key role in the future energy revolution and industrial revolution.HDR reservoir is a rock with low fluid content,a buried depth of3 ? 8km and a temperature of 200-350 ?,related to active semi-solid and solid rheological structure.The research on the genesis,exploration,evaluation and development of HDR is very weak,and many fields are still blank.There is an urgent need to strengthen the research on HDR reservoirs' formation mechanism and distribution law.The change of underground thermal structure caused by water injection fracturing in geothermal well,dynamic monitoring in the production stage and subsequent prospect evaluation.Geophysical methods based on the characteristics of reservoir physical parameters,such as low-frequency electromagnetic,gravity,and spontaneous potential,are necessary technical means in exploring,developing,and monitoring HDR reservoirs.The application research of conventional geophysical methods mainly focuses on reservoir description and HDR target evaluation.There is little research on the formation mechanism and temporal and spatial variation law of HDR thermal reservoir,especially the monitoring of reservoir intrinsic parameters such as fracture extension,migration and distribution of injected fluid and reservoir permeability before and after artificial fracturing in the process of mining.In this paper,the time-lapse monitoring of the HDR reservoir is mainly carried out through various geophysical methods.According to the reservoir gravity anomaly,electrical parameters(resistivity,impedance phase)and velocity changes,the timelapse monitoring of the critical parameters of HDR reservoir is carried out from different scales and physical laws,and the dynamic model of reservoir parameters is established.On this basis,the machine learning method is used to organize and classify the existing time-lapse data of HDR reservoir,modify and calculate the reservoir dynamic model,and predict the variation law of heat source parameters such as reservoir terrestrial heat flow and geothermal gradient.Its primary contents include 1 The forward coupling calculation of temperature field,and pressure field through multi-scale fracture network constitutes an essential premise for underground imaging and characterization of multi geophysical methods.Among them,the crack network is calculated using the governing equation.The medium space is matched by the connection function,which effectively solves the problems of insufficient treatment of small cracks in the fine grid processing of the existing calculation software.2.Realize the conversion between the heat source and geophysical parameters(density,resistivity and velocity parameters)through variation empirical function and machine learning statistical method.3.Calculate the parameters through various geophysical forward modelling methods,obtain geophysical response signals of different scales,and explain the mechanism of the geothermal reservoir in different aspects,including the interpretation of natural potential according to the kinematic characteristics of water migration.The difference in the electrical structure is caused by the change of underground rock and seepage water.The anisotropic phase tensor method locates the fracture trend and tendency development produced in underground fracturing.The gravity method explains the density change in the process of water injection fracturing one by one and then gives a systematic geophysical geothermal response standard evaluation,including the differences and laws of the deep heat source,heat conduction channel and caprock influence of the conventional thermal reservoir in each geophysical response change.4.Combined with multiple geophysical signals and geological information,the learning framework is constructed through a machine learning neural network method to train the earth heat flow in Songliao Basin.The logarithm results,such as the deep learning sensitivity interpretation method,are used for uncertainty analysis to finally realize the prediction of underground thermal response in future development and production.The research content of this paper can provide efficient and reliable geophysical methods,technologies and prediction basis for the development and evaluation of HDR geothermal resources and guide the subsequent development of HDR.