中文摘要：

差异演化（differential evolution，DE）是Storn和Price（Technical ReportTR-95-102，International Computer Institute，Berkely，1995）提出一种基于个体差异重组思想的演化算法，适合于求解连续空间的最优化问题．和其它演化算法相比，差异演化算法在求解非凸、多峰、非线性函数优化问题时表现出极强的稳健性，且在同样的精度要求下，算法收敛的速度快,但过早收敛和陷入局部最优是包括差异演化在内的演化算法面临的一个重要问题,提出一种基于Boltzmann生存机制的双子代竞争差异演化算法,为避免算法过早收敛，利用交叉操作生成两个新个体以增加群体多样性，然后与父代个体竞争形成子代个体．在选择操作中引入Boltzmann机制，以一定概率接受较差解，使算法能跳出局部最优，最终达到全局最优解．利用Brest et al．（Evdutionary COmputation，2006，10：646-657）中的21个测试函数，分别与标准DE算法、jDE算法进行性能比较．实验结果表明，该算法的平均性能值、最优性能值以及最优解质量都优于标准DE算法和jDE算法．

英文摘要：

Differential evolution （DE） that was developed by Storn and Price（Technical Report TR-95-102, International Computer Institute, Berkely, 1995） is one of the most successful evolution algorithms for the global continuous optimization problem. DE utilizes the mutation and recombination operators as search mechanisms, and the selection operator to direct the search towards the most promising regions of the solution space. Differential evolution algorithm is much more robust and was quicker convergence rate for nonlinear, multimodal functions than other evolution algorithms. However, as a particular instance of evolution algorithm, although it is simple and powerful for optimizing continuous functions, differential evolution algorithm is still faced with premature convergence and to get involved in local optimization problems just like other evolution algorithms. In this paper, adifferential evolution algorithm with double trial vectors based-on boltzmann mechanism （boDE） is presented. Two trial vectors are created by recombination to increase colony diversity and avoid premature convergence. These vectors compete with the parent individual to produce the next generation. Moreover, we introduce the boltzmann mechanism into the selection operator. This mechanism makes some not bad individuals accepted and makes the algorithm depart from the local optimization. The simulations have been finished for twenty-one benchmark functions with three evolution algorithms （SDE, jDE and boDE）. Experimental results indicate that the proposed algorithm is efficient and feasible. It is superior to other related methods such as SDE, jDE both on the quality of solution and on the on-line and off-line performance.