中文摘要：

将反馈分析问题归结为约束最优化问题,采取最为直接的正演反分析的求解方案,即对结构进行正向数值分析以实现约束条件,利用人工神经网络进行参数反演以满足目标函数,从而建立基于神经网络的反馈分析方法,并给出详细实现步骤。该方法分析过程简单,通用性强。以溪洛渡水电站左岸地下厂房洞室群开挖过程的反馈分析为例,根据工程施工期位移监测资料,以三维连续介质快速Lagrange分析程序FLAC3D作为数值分析软件,建立神经网络数值反馈分析系统。围岩位移反馈分析成果与实测数据吻合,后期开挖的围岩变形、应力以及支护结构受力等的预测成果合理,这可作为溪洛渡厂房洞室群的围岩稳定性评价的可靠基础,也证明该方法是解决大型复杂工程监测反馈分析问题的有效途径,可能得到进一步的推广和应用。

英文摘要：

The feedback analysis of underground powerhouse caverns excavation of large-scale hydropower stations attracts more and more attention recently.The effective feedback analysis method is very crucial to such problem.A direct solution scheme based on artificial neural network（ANN） is suggested considering that the feedback analysis is virtually a constrained optimization problem.It mainly consists of analyzing the structure numerically to build the constraints and obtaining the input parameters by ANN to fulfill the objective function.The steps of this method are also illustrated in detail,from which can find that the method is simple but universal.The ANN based feedback analysis method can avoid the deficiency of the traditional mathematical programming and improve the effectiveness and the convergence of feedback analysis.Taking the underground powerhouse caverns excavation of Xiluodu hydropower station in China for examplet,he fast Lagrangian analysis of continua,FLAC3D,is used to establish the ANN based feedback analysis system.The feedback analysis with the monitoring deformation data during excavation is conducted.The surrounding rock displacements obtained by feedback analysis agree with the measured displacements.The predicted deformations and stresses of caverns and the loads on the supporting structures in the next excavation are also reasonable.These not only help to evaluate the stability of the powerhouse caverns of Xiluodu hydropower station but also show that the method proposed is valid and applicable for complex monitoring feedback analysis problems;and the method may get extensive application in engineering.