面向定制产品进化设计的优势种群产生策略

为了提高个性化定制产品进化效率,提出了一种在相似实例等距布种的基础上进行等位基因多样化阈值控制的优势种群产生策略。阐述了个性化定制产品优势种群的特征,认为精英个体应同时具有定制需求宏观上的相似性和种群基因微观上的多样性两大特征。在产品基因建模的基础上,采用夹角余弦作为聚类统计量,提取具有相似定制基因组的历史实例作为优势种群的初始解;基于加权的闵科夫斯基距离,对初始解成员进行等距采样,形成差异化精英个体候选队列,以实现种群的均匀布种;采用杰卡德距离度量候选个体承载的基因组与当前初始基因库两者间差异度,提出了等位基因阈值矩阵,构建了多样化等位基因生成算法。给出了产品适应度评价函数,基于评价反馈对优势度低的初始种群进行改进。以桥式起重机为例,对提出的策略进行了验证。实验结果表明,该策略能同时兼顾进化效率和进化质量,有助于避免早熟收敛现象的发生,适合个性化定制产品的进化设计。

Generating Strategy of Superior Population for Customization Products Evolutionary Design

The development of initial population generation strategy was briefly reviewed. In order to realize the efficient evolution design of customization products, a reasonable model and algorithm was proposed to produce the superior population. Characteristics of the superior population for customized products were discussed. The two remarkable features were that elite individuals should have the similarity of customized requirements and initial genes should be diverse. Based on product gene model, historical product design cases with similar custom genome were obtained by using the angle cosine as clustering statistic, and these cases were used as the initial solution for superior population. Elite individual candidate queue was established by sampling the initial solution members with equal weighted Minkowski distance to achieve uniform population distribution. Difference between the elite individuals genome and the current initial gene library was measured by Jaccard distance. Allele threshold matrix was defined and a diversity allele generation algorithm was built to achieve the diversity of initial population genes. The product fitness evaluation function was established and the dominance of the initial population could be improved according to the feedback result of the evaluation function. The strategy and method were effective in an instance of the product called bridge crane. The experimental results showed that the differentiation of individual queues and the control of multiple gene thresholds could help to avoid premature convergence and obtain high fitness solutions, the proposed approach could take into account both the evolutionary efficiency and the evolutionary quality and increase the success rate of evolution. The strategy was suitable for customized product evolutionary design.