【摘要】 很多实际工程和科学问题都是计算费时的多目标优化问题，这类问题中每个候选解的评价往往都非常费时，因此仅允许使用少量真实评价。论文采用离线数据驱动的进化算法求解计算费时多目标优化问题，以期节省优化时间。论文通过训练代理模型来估计候选解的收敛性，采用最近邻样本估计候选解与参考向量的关联关系，减少了使用目标估值计算候选解与参考向量夹角大小所产生的误差累积。使用DTLZ测试集验证论文算法的有效性，论文算法与离线数据驱动的优化算法MS-RV以及三个经典在线数据驱动优化算法进行对比，实验结果表明论文提出的算法在保证性能的前提下，可以减少使用真实的评价次数。更多还原

【Abstract】 Many real-world engineering and science problems are computationally expensive multi-objective optimization problems,in which the evaluation of each candidate solution is time-consuming. Thus,only a small number of real objective evaluations are allowed. In this paper,an offline data-driven evolutionary algorithm is proposed for solving computationally expensive multi-objective optimization problems,which is expected to save the time of optimization. In the proposed method,the surrogate model is trained to approximate the convergence performance of a candidate solution. The reference vector that a candidate solution is associated and determined by the nearest sample of the candidate solution. And accordingly,it is expected to reduce the accumulated errors resulting from calculating the angle between the approximated objective values and the reference vector. The DTLZ test set is used to verify the effectiveness of the proposed method. The proposed method is compared with an offline data-driven optimization algorithm named MS-RV and three classic online data-driven optimization algorithms. Experimental results show that our proposed method can reduce the number of real evaluations without deteriorating the performance.更多还原