The problem of optimally designing an interplanetary trajectory for a space mission is considered in this paper. To tackle the extreme non-linearity of the search space, a case learning-based differential evolution algorithm, named CLDE, is proposed. It stores successful control parameters (scaling factor and crossover possibility) and retrieve the available reference information according to a geographic similarity in each generation. To depart from the basin of attraction of a local optimum, CLDE will give up learning from the successful cases once no better offsprings have been obtained within a certain number of generations and generate new control parameters. Two versions of CLDE have been developed, for global optimization (G-CLDE) and local optimization (L-CLDE), respectively. Their performance has been tested on GTOP benchmarks and real mission design. Experimental results show that G-CLDE performs better than related algorithms, including PYGMO algorithms and recently published L-SHADE variants. L-CLDE can improve upon the best known solution for the Messenger benchmark (full version). By connecting G-CLDE and L-CLDE together, CLDE finds promising results in acceptable computational time on the GTOP benchmark.

本文考虑了为太空任务优化设计行星际轨道的问题。为了解决搜索空间的极端非线性问题，提出了一种基于案例学习的差分进化算法，称为CLDE。它存储成功的控制参数（比例因子和交叉可能性），并根据每一代中的地理相似性检索可用的参考信息。为了摆脱局部最优吸引域，一旦在一定数量的世代中没有获得更好的后代并产生新的控制参数，CLDE将放弃从成功案例中学习。已经开发了两种版本的CLDE，分别用于全局优化（G-CLDE）和局部优化（L-CLDE）。他们的性能已经通过GTOP基准测试和实际任务设计进行了测试。实验结果表明，G-CLDE的性能优于相关算法，包括PYGMO算法和最近发布的L-SHADE变体。L-CLDE可以改进Messenger基准（完整版）这一最著名的解决方案。通过将G-CLDE和L-CLDE连接在一起，CLDE可以在GTOP基准上可接受的计算时间内找到有希望的结果。