【作者】 董华松；

【导师】 黄文辉；

【作者基本信息】 中国地质大学（北京） ， 矿产普查与勘探， 2014， 博士

【摘要】 在常规资源匮乏情况日益加重的今天，清洁、绿色、可再生能源的使用成为新一轮能源调整的重点，受到世界各国的重视。其中，地热资源以其资源储备量雄厚，方便开发利用，可持续性，而受到人们的高度关注。我国油田的油井伴生地热资源丰富，目前主要的采热手段为注水采热，二氧化碳羽流系统作为一种新的采热方法，以其更强的采热效果和对于二氧化碳地质封存的潜在能力，一经提出就成为研究的热点。但不论是哪种注入流体采热的方式，部分储层参数和注入条件对于采热效果的影响和原因，以及智能算法对采热效果的回归预测，尚未见详细报导。本文首先利用数值模拟的方法，分别对注入水和超临界二氧化碳开采油井地热进行建模，通过改变相应的储层参数和注入条件，分析其对采热效果的影响，并根据影响的时间、大小、方式分析成因。根据数字模拟的结果，找出对于注入不同流体采热的相应影响因子，通过智能算法中的深度学习方法对多井采热、单井采热和采热趋势变化进行回归预测。通过上述模拟和回归预测实验，得到以下几点的认识：1.通过对注水采热和注入超临界二氧化碳采热系统的数值模拟，可以看出后者在井下具有超强的流动性，虽然热晗值低于水，但总体的采热效果将近注水采热的一倍。2.对注水地热开采来说，储层温度、注水量、注水速度、储层岩石比热容、注入井和生产井井径、流体注入温度的升高，会提高注水采热的效果；储层初始盐度的升高会降低注水采热的影响效果；储层岩石渗透率、储层岩石导热系数和储层压力对采热效果影响不大。3.对于注入超临界二氧化碳采热来说，储层温度、储层岩石渗透率、注入量、注入速度、储层岩石导热系数、储层岩石比热容、初始注入温度的升高能够提高采热效果。该方法对储层压力变化非常敏感，会使得采热效率发生很大变化。储层初始盐度在水和二氧化碳两相流阶段的升高会降低采热效果，注入井和生产井井径对于采热的作用需根据地层压力来确定。4.经过实验和与传统人工神经网络的比较，采用DBNs+SVM模型，对多井采热、单井采热和采热变化趋势进行回归预测的效果相对较好，能够从样本数据的抽象特征中找出规律，并进行回归预测，预测数据与原始数据的拟合效果良好。从实验结果来说，该方法比较灵活，能够自适应的建立网络模型，适合于储层条件和初始注入条件的随时变化，相对于传统数值模拟软件来说，具有一定的优势。更多还原

【Abstract】 With scarcity of conventional resources becoming more and more serious, the usage ofclean, green and renewable energy becomes the focus of next energy construction adjustment,which obtains attention all over the world. Among those new energy resources, the geothermalresource become a real priority because of its high reserves, convenient utilization and strongsustainability. Geothermal exploitation means of oil wells at present are water injection andCO2plume system. The latter, as a new method, get the attention of industry because of itsability of strong heat getting and potential advantage for CO2sequestration. But no matter whatkind of heating method, thermal effect and cause of some reservoir parameters and injectionconditions, and regression prediction with intelligent algorithm for thermal effect, has not beenreported.The thesis was written as follows: firstly, modeling for geothermal development of oilwells with injecting water and supercritical carbon dioxide with numerical simulation method;by changing the corresponding reservoir parameters and injection conditions, we computed itsinfluences on geothermal development of oil wells and analyzed its causes according to timepoint, degree and style of influences; secondly, according to the simulation result, we find outthe corresponding influencing factors of different fluid for geothermal development of oil wells;and then using deep learning as the method to predict thermal efficiency of multi-wells model,single well model and change trend of thermal efficiency. Through the experiment of numericalsimulation and intelligent regression, the following conclusions were reached：1. Through numerical Simulation of water and supercritical carbon dioxide injectiongeothermal exploitation system, we can see that the latter has superior buoyant, thoughenthalpy is lower than the water, its overall thermal efficiency nearly doubled to water.2. Improvement of parameters like reservoir temperature, injection volume, injectionrate, specific heat capacity of reservoir rock, diameter of injection well and production well,fluid injection temperature will increase thermal efficiency of water injection; Improvementof initial salt concentration of reservoir will reduce thermal efficiency of water injection; pressure, permeability and thermal conductivity of reservoir rock has little effect onthermal efficiency of water injection.3. Improvement of parameters like reservoir temperature, permeability of reservoirrock, injection volume, injection rate, thermal conductivity of reservoir rock, specific heatcapacity of reservoir rock, initial temperature of injected fluid can increase thermalefficiency of scCO2injection. Supercritical carbon dioxide injection geothermalexploitation system is very sensitive to the change of reservoir pressure, which will lead itto change a lot. Increasing initial level of reservoir salinity will reduce thermal efficiency ofscCO2injection in two-phase flow of water and CO2. Affection of diameter of injection andproduction well on thermal efficiency of scCO2injection should be depended on reservoirpressure.4. Deep belief network can complete the high dimensional feature extraction. Throughthe comparison between it and traditional artificial neural network, it was found that DBNshad been effective enough in regression prediction for multi-wells model, single-wellmodel and change trend of thermal efficiency, and could find rules from abstractcharacteristics of sample data, then predict thermal effect. The result proved thatforecasting data fitted well with the original data, but the side effect is time-consuming ifthe amount of hidden layer nodes was too big. All in all, intelligent prediction forgeothermal development of oil wells with injecting water and scCO2is more flexible, andable to build an adaptive network model, which is suitable for changes of initial injectionconditions and reservoir conditions at any time. Compared with the traditional numericalsimulation software, this method has certain advantages in some aspects, and has valuablereference for the future production.更多还原