基于迁移学习灰支持向量回归机的交互式进化计算

Interactive evolutionary computation based on transfer learning grey support vector regression

针对机器感知评价和种群进化,提出基于迁移学习灰支持向量回归机的个体适应值预测方法和聚类进化策略.通过共享用户已评价个体适应值学习模型与部分未评价个体适应值学习模型,实现知识模型差异最小化.建立具有迁移学习能力的灰支持向量回归机模型,预测未评价个体适应值.基于聚类子集计算个体平均距离,并设计选择算子和交叉算子,扩大子代搜索区域,增强种群多样性.基于上述策略,采用NSGA-II范式实现交互式进化计算.最后,分析算法时间复杂度,表明算法可提高评价精度,并克服局部收敛问题.将该算法应用于室内灯光调色问题,验证所提出方法的有效性.

This paper proposes a fitness prediction method based on transfer learning grey support vector regression with the clustering evolution strategy aiming at machine perception evaluation and population evolution, which can minimize knowledge model differences by sharing the user learning model of individual evaluation fitness and the part of unevaluated individuals, and can forecast unevaluated individuals fitness. A selection and crossover operator based on clustering individual average distance are proposed, which can expand offspring search area and enhance population diversity. Based on above strategies, the interactive evolutionary computation applied to NSGA-II is proposed. Finally,the analysis of the time complexity indicates that the proposed method can improve evaluation precision, and overcome the local convergence problem. The method is applied to interior light toning optimization problem, and its outstanding performance is experimentally demonstrated.