Stochastic characterization of discrete fractures in rock by hydraulic and tracer tomography

裂缝岩中含有丰富的含水层，是增强型地热系统（EGS）的目标。这些学科共同面临的一个主要挑战是与流动和输送过程相关的结构特征或裂缝系统的适当表征。表征越准确，数值模型对过程的模拟就越可靠。数值模拟技术的最新发展令人印象深刻，在计算效率、逼真度、高分辨率和耦合模拟方面的能力不断增强。尽管如此，与此相关的数值模型的数据饥渴几乎没有得到现场测量的满足。特别是每个地点的独特特征，如裂缝几何形状，需要调整现场调查技术。即使对一个地点进行了充分的调查，也需要将测量数据整合到模型中的方法。该项目建议使用层析成像井眼测试，注入水和示踪剂，以识别和表征与流动和输送相关的裂缝。通过结合在多个井眼中进行的跨井多级测试的见解，可以促进裂缝几何形状的重建。一个主要的新元素是通过一种通用的贝叶斯方法来反演记录的层析信号，该方法可以迭代调整裂缝的方向、长度和裂缝密度（逆模型）。这与离散裂缝网络（正演模型）的灵活快速的数值实现和模拟相结合。在初步结果的鼓舞下，本文进一步发展了层析离散裂缝反演方法，用于在二维和三维系统中进行裂缝概率的鲁棒估计。利用虚拟井眼试验的综合数据，以及压裂岩石原位实验的压力和热示踪数据，对该方法进行了阐述和论证。

Fractured rocks host productive aquifers and they are the target for enhanced geothermal systems (EGS). A major challenge shared by these subjects is the appropriate characterization of structural features or fracture systems that are relevant for flow and transport processes. The more accurate the characterization, the more reliable the processes can be simulated by numerical models. Recent developments in numerical modelling techniques are impressive, with growing capabilities in computationally efficient, realistic, high-resolution and coupled simulation. Still, the associated data hunger of numerical models is barely fed by available field measurements. Especially features unique for each site, such as fracture geometries, require attuned site investigation techniques. And even if a site is well investigated, methods are needed for integrating measured data in a model. This project proposes the use of tomographic borehole tests with water and tracer injection to identify and characterize fractures relevant for flow and transport. By combining the insight from cross well multi-level tests in several boreholes, the reconstruction of fracture geometries is facilitated. A major novel element is the inversion of recorded tomographic signals by a versatile Bayesian approach that adjusts iteratively fracture orientations, lengths and fracture density (inverse model). This is combined with a flexible and fast numerical implementation and simulation of the discrete fracture network (forward model). Motivated by promising preliminary results, the proposed tomographic discrete fracture inversion approach is here further developed for robust estimation of fracture probabilities in two- (2D) and three-dimensional (3D) systems. It is elaborated and demonstrated utilizing synthetic data from virtual borehole tests, as well as by means of pressure and heat tracer data from in-situ experiments in fractured rock.