Aiming at the phenomenon of discrete variables whic h generally exists in engineering structural optimization, a novel hybrid genetic algorithm (HGA) is proposed to directly search the optimal solution in this pape r....Aiming at the phenomenon of discrete variables whic h generally exists in engineering structural optimization, a novel hybrid genetic algorithm (HGA) is proposed to directly search the optimal solution in this pape r. The imitative full-stress design method (IFS) was presented for discrete struct ural optimum design subjected to multi-constraints. To reach the imitative full -stress state for dangerous members was the target of IFS through iteration. IF S is integrated in the GA. The basic idea of HGA is to divide the optimization t ask into two complementary parts. The coarse, global optimization is done by the GA while local refinement is done by IFS. For instance, every K generations, th e population is doped with a locally optimal individual obtained from IFS. Both methods run in parallel. All or some of individuals are continuously used as initial values for IFS. The locally optimized individuals are re-implanted into the current generation in the GA. From some numeral examples, hybridizatio n has been discovered as enormous potential for improvement of genetic algorit hm. Selection is the component which guides the HGA to the solution by preferring in dividuals with high fitness over low-fitted ones. Selection can be deterministi c operation, but in most implementations it has random components. "Elite surviv al" is introduced to avoid that the observed best-fitted individual dies out, j ust by selecting it for the next generation without any random experiments. The individuals of population are competitive only in the same generation. There exists no competition among different generations. So HGA may be permitted to h ave different evaluation criteria for different generations. Multi-Selectio n schemes are adopted to avoid slow refinement since the individuals have si milar fitness values in the end phase of HGA. The feasibility of this method is tested with examples of engineering design wit h discrete variables. Results demonstrate the validity of HGA.展开更多
文摘Aiming at the phenomenon of discrete variables whic h generally exists in engineering structural optimization, a novel hybrid genetic algorithm (HGA) is proposed to directly search the optimal solution in this pape r. The imitative full-stress design method (IFS) was presented for discrete struct ural optimum design subjected to multi-constraints. To reach the imitative full -stress state for dangerous members was the target of IFS through iteration. IF S is integrated in the GA. The basic idea of HGA is to divide the optimization t ask into two complementary parts. The coarse, global optimization is done by the GA while local refinement is done by IFS. For instance, every K generations, th e population is doped with a locally optimal individual obtained from IFS. Both methods run in parallel. All or some of individuals are continuously used as initial values for IFS. The locally optimized individuals are re-implanted into the current generation in the GA. From some numeral examples, hybridizatio n has been discovered as enormous potential for improvement of genetic algorit hm. Selection is the component which guides the HGA to the solution by preferring in dividuals with high fitness over low-fitted ones. Selection can be deterministi c operation, but in most implementations it has random components. "Elite surviv al" is introduced to avoid that the observed best-fitted individual dies out, j ust by selecting it for the next generation without any random experiments. The individuals of population are competitive only in the same generation. There exists no competition among different generations. So HGA may be permitted to h ave different evaluation criteria for different generations. Multi-Selectio n schemes are adopted to avoid slow refinement since the individuals have si milar fitness values in the end phase of HGA. The feasibility of this method is tested with examples of engineering design wit h discrete variables. Results demonstrate the validity of HGA.
文摘针对高斯混合模型(Gaussian mixture model,GMM)参数选取效率较低的问题,提出了一种在基于GMM的轨迹模仿学习表征中综合求解GMM参数估计的方法.该方法基于多中心聚类算法中的最大最小距离算法改进kmeans算法,得到最优初始聚类中心,并基于贝叶斯信息准则(Bayesian information criterion,BIC)通过遗传算法优化求解,同时获取GMM的4个重要参数.该方法通过提高划分初始数据集的效率,在优化初始聚类中心基础上确定混合模型个数,有效地避免了因为初值敏感而导致的局部极值问题.通过多组仿真实验验证了该方法的有效性.
