离散裂隙随机网络模型中结构面的分布特征仅代表其统计规律，并非地质成因、结构模式、构造形迹的客观反映，以致后续力学计算及渗流模拟结果可信度偏低。针对国家高放废物处置库甘肃北山预选区出露良好的花岗岩体，基于实测结构面统计规律、内在成生关系及水力联系，在前期纯随机模型关键部位修正特征参数，并融入人工辨识的确定结构模式而构建“随机-确定”耦合模型。图形和渗流模拟两种定量检验结果表明：耦合模型结构面数量更接近客观实际，模型准确度提升约48.8%；耦合模型渗流路径与流量较之前更显客观真实，模拟结果与试验数据的接近程度比随机模型大约1/3。另外，不同孔位渗流结果显示：相比于随机结构面，确定性结构面对区域渗流控制作用更加明显，在渗流模拟中扮演更为重要角色。此耦合模型有益于拓宽结构面网络模拟的发展方向。

Since stochastic discrete fracture network (DFN) model can only consider statistical properties of discontinuities, it is difficult to incorporate other significant factors, such as geological genesis, structural pattern and feature. To fully characterize the true DFN of rock mass, non-negligible biases are introduced into the subsequent mechanical and seepage calculations based on a pure stochastic model. The Beishan granite, a well-studied rock type, from a Chinese high-level radioactive waste repository, is selected for the analysis. A new stochastic-deterministic DFN model is built up for Beishan granite in terms of the statistical properties of measured discontinuities, the inherent hierarchy of discontinuities with different scales, and their hydraulic interconnections. The deterministic part of this model is implemented after combining the unbiased stochastic model with manually identified structural patterns. Model validations are conducted using the graphical comparison and seepage simulation techniques. The results show that the amount of discontinuities in the new model is consistent with the measured data, with an increase in model accuracy by 48.8%. The flow path and flux calculated by the new model are more realistic and more consistent with testing data, with approximately a one-third increase in consistency compared with the traditional model results. In addition, in-situ seepage-pressure tests are conducted on different boreholes. Testing results show that the deterministic discontinuities have great influence on the seepage simulation compared with the stochastic ones. The regional seepage is primarily controlled by these deterministic discontinuities. The developed new model is beneficial for the future DFN simulations.