作为非常规油气储层改造的重要环节, 准确描述压裂缝网空间分布是评估水力压裂效果和计算油气采收率的基础。本文采用了水力压裂流固耦合数值模型表征压裂缝网动态扩展行为并同步输出对应的井口压力响应, 在此基础上集成了可逆跳跃马尔可夫链蒙特卡罗 (RJ-MCMC) 算法与并行计算架构, 构建了一种基于现场施工压力曲线实时反演压裂缝网几何形状的贝叶斯变维反演方法。该方法以缝网内离散裂缝单元 (几何特征与空间分布) 作为随机变量, 基于前一时刻输出的缝网几何形态和该时刻裂缝扩展在不同线程上同时进行随机采样, 结合物理模型与观测数据, 采用Metropolis-Hastings-Green (MHG) 准则筛选随机采样结果, 从而快速生成符合实测井口压力数据的缝网样本集合。随后, 采用贪心算法 (Greedy Algorithm) 选择压力匹配度最高的缝网几何形态为反演结果, 同时优化下一时间步的缝网随机参数以实现后续时间步的高效反演。最后, 为验证反演方法的可行性, 利用西南页岩气H井4段压力数据反演压裂缝网, 并与微地震监测结果对比。结果表明, 反演结果与监测的缝网几何形态在时间和空间上较为一致。必须指出的是, 虽然该反演方法在压力数据选取等方面仍存在改进之处, 但是它无疑为压裂施工缝网几何形态实时评估提供了一个新手段。

As a critical component in the stimulation of unconventional oil and gas reservoirs, the accurate characterization of the spatial distribution of hydraulic fracture networks forms the basis for evaluating fracturing effectiveness and calculating hydrocarbon recovery. In this study, a coupled fluid-solid numerical model for hydraulic fracturing is employed to characterize the dynamic propagation behavior. of fracture networks, while simultaneously capturing the corresponding wellhead pressure responses. Based upon this, a Bayesian trans-dimensional inversion method is developed by integrating the Reversible Jump Markov Chain Monte Carlo (RJ-MCMC) algorithm with a parallel computing framework, enabling real-time inversion of fracture network geometry based on field injection pressure curves. In this approach, discrete fracture elements within the network—characterized by their geometric properties and spatial distribution— are treated as stochastic variables. Given the independence of fracture growth, random sampling is carried out in parallel across multiple threads. By combining the physical model with observational data, the Metropolis-Hastings- Green (MHG) criterion is applied to screen the sampling results, efficiently generating a set of fracture network samples that align with the measured wellhead pressure data. Subsequently, a Greedy Algorithm is applied to select the fracture geometry with the best pressure-matching accuracy as the inversion result and to optimize the fracture network parameters for the subsequent time step, thereby improving inversion efficiency. Finally, to verify the feasibility of the proposed method, pressure data from Stage 4 of Well H in a shale gas field in southwestern China were used to invert the fracture network geometry, and the results were compared with microseismic monitoring data. The comparison demonstrates strong consistency in both time and space between the fracture geometries derived from inversion and those from microseismic monitoring. Although aspects of the inversion method, such as pressure data selection, still require improvement, the proposed approach undoubtedly provides a novel solution for the real-time evaluation of fracture geometries during hydraulic fracturing operations.