基于并行改进遗传算法的拱坝位移反分析

Inverse analyses of arch dam displacements using improved parallel genetic algorithm

引入小生境技术和自适应杂交变异概率方法,基于并行有限元程序,给出了适合推求拱坝和地质力学参数的位移反分析并行遗传算法,并编制了相应的程序,利用地质力学模型试验数据,对溪洛渡拱坝进行了位移反分析,得到了和试验相一致的坝体混凝土和地基岩体的力学参数。结果表明,该算法可以有效解决简单遗传算法的早熟收敛问题,收敛效率得到明显提高。当采用16个CPU进行并行计算时,可以达到42%的计算效率,表明该算法适用于拱坝这样复杂的三维结构的位移反分析,可以大大减少拱坝位移反分析的时间。

Classic genetic algorithms require a large number of calculations for back analyses of arch dam displacements with many false solutions. Based on niche technology adaptive crossover and mutation probabilities method improves the genetic algorithms results. A parallel genetic algorithm, which uses the adaptive probabilities, was developed for back analyses of arch dam displacements. The displacement analysis results for the Xiluodu arch dam agreed well with displacements measured in model experiments. The results show that the method avoids false solutions that often occur with simple genetic algorithms and has good convergence. The parallel computation efficiency was 43% with 16 CPUs. Therefore, the algorithm is suitable for back analyses of complex 3-D structures like arch dams.