中文摘要：

面板堆石坝堆石体力学参数反演优化问题是一个多变量、多约束的混合非线性规划问题，当正演过程用神经网络模拟器替代后，高效快捷的优化算法成为解决问题的关键。提出一种用以解决这一复杂优化问题的混合算法——混沌直接搜索粒子群（CHPSO—DS）算法。在改进的算法中，首先结合混沌优化思想对粒子群进行初始化，减轻粒子初始位置的选择对算法优化性能的影响；利用直接搜索法克服了粒子群算法后期搜索效率降低的缺陷，提高算法局部搜索能力。为证明该算法的优越性，同时将该算法与遗传算法（GA）用于水布垭面板堆石坝堆石体力学参数的位移反分析计算中。实践证明，利用CHPSO—DS算法搜索时能快速收敛到全局最优解，且算法具有较强的鲁棒性；两算法对比结果也表明，不论是优化精度还是收敛时间，CHPSO—DS算法都较GA有明显提高。最后利用CHPSO—DS算法反演的堆石体力学参数进行测点沉降预测，结果表明各个测点的计算位移值与监测值吻合较好，说明CHPSO—DS算法在复杂岩土工程位移反分析中具有良好的实际应用价值，值得进一步研究和推广。

英文摘要：

Rockfill parametric inversion is a multi-variable and multi-constraint nonlinear optimization problem. When the finite element analysis is conducted by the simulator of neural network, highly efficient optimizing algorithm is the problem-solving key. A modified particle swarm optimization algorithm, chaotic particle swarm optimization with direction search（CHPSO-DS） algorithm, is provided to solve the complex problem. In the CHPSO-DS algorithm, the particle is initialized with chaos optimization method in its sub-area, which reduces the influence caused by initial position of particle, and then the local search capability of the algorithm is increased by direct search method. For comparison, the CHPSO-DS algorithm and genetic algorithm are used to back analyze parameters of Shuibuya concrete face rockfill dam on the basis of measured displacements. The results show that the CHPSO-DS algorithm can converge quickly and is very robust, and it takes shorter time compared with genetic algorithm in a same precision level. Those show that CHPSO-DS algorithm is very excellent. At last, the calculated mechanical parameters are used to forecast the settlements of the monitoring points of Shuibuya concrete face rockfill dam. Forecasted values are in good agreement with the measured values, which indicates that the CHPSO-DS algorithm can be well applied to the displacement back analysis in geotechnical engineering.