【摘要】 在岩体工程领域，岩石的峰后强度对工程结构的稳定性至关重要。为了精确地描述岩石在峰后阶段的力学行为，提出了基于连续参数动态修正的耦合技术框架。利用Python建立试验数据驱动的实时反向传播算法，通过FLAC^3D内嵌接口实现莫尔—库伦本构参数演化函数的多线程动态迭代，突破传统固定参数修正法的精度瓶颈，开发了基于实时数据同化的多参数逆向优化引擎，构建了应变梯度自适应的三维本构场动态迭代模型。研究结果表明：原始模型的应力水平始终高于实际值，而修正模型的应力模拟值与试验值幅值匹配，高估趋势完全消除。本研究为岩石参数精度缺失的优化提供了新思路，研究成果可为岩土工程设计提供科学依据和指导。更多还原

【Abstract】 In the field of rock mass engineering, the precise simulation of the post-peak mechanical behavior of rocks is essential for ensuring engineering safety and for the prevention and management of disasters. To overcome the limitations associated with fixed parameter approaches in conventional Mohr-Coulomb models for simulating post-peak failure stages, this study introduces an innovative experimental data-driven multiparameter dynamic collaborative correction method. Initially, by integrating Python with the FLAC^3D platform, we developed a real-time backpropagation algorithm alongside a three-dimensional constitutive field dynamic iteration model. This framework facilitates the multi-threaded collaborative optimization of parameters, including cohesion, internal friction angle, and dilatancy angle, via embedded interfaces. Subsequently, utilizing strain gradient adaptive theory, we devised a real-time data assimilation engine capable of dynamically adjusting constitutive parameters through cyclic correction mechanisms. This approach effectively addresses the modeling challenges posed by the nonlinear coupling effects inherent in traditional static segmentation methods, which exhibit errors exceeding 15%.During the validation process, a numerical model for uniaxial compression, with dimensions of 50 mm×50 mm×100 mm and comprising 2 541 mesh elements, was utilized. A Python script was employed to dynamically invoke the s. stress（） [2] [2] function in FLAC^3D, allowing for the extraction of stress fields. This process initiated multi-parameter collaborative corrections whenever the experimental data surpassed a deviation threshold of Δσ =0.01 MPa. The experimental findings demonstrate that the dynamically corrected model effectively captured the post-peak strain-softening behavior of the rock. The stress levels predicted by the original model consistently exceeded the actual values, whereas the stress simulations from the corrected model aligned closely with the experimental values, thereby completely mitigating the trend of overestimation. This study offers novel insights into optimizing the accuracy of rock parameter estimation and provides a scientific basis and guidance for geotechnical engineering design.更多还原