论文部分内容阅读
基于梯度信息的线性搜索法具有快速的收敛性,但易陷入局部最优。当优化目标不可解析时,基于梯度信息的算法便不易应用。多目标进化算法以其优秀的全局特性广泛地应用于多目标优化问题,但其算法比较耗时,收敛速度慢。对此,本文提出一种基于进化梯度搜索的多目标混合算法。首先,结合单目标优化中的爬山算法与进化梯度搜索法,得到一种多目标局部搜索算法。其次,在算法前期采用适应度概率策略选择个体进行局部搜索。最后,在非支配集个体数达到种群个体数后,应用多目标进化算法保证其分布性。通过ZDT系列测试函数验证并与NSGA-II及EGS-NSGA-II混合算法比较,结果显示本算法具有更好的全局性及收敛快速性。
The linear search method based on gradient information has fast convergence, but it is easy to fall into the local optimum. When the optimization goal can not be solved, the algorithm based on gradient information is not easy to apply. Multi-objective evolutionary algorithms are widely used in multi-objective optimization problems for their excellent global characteristics, but their algorithms are time consuming and slow in convergence. In this regard, this paper presents a multi-objective hybrid algorithm based on evolutionary gradient search. First, a multi-objective local search algorithm is obtained by combining the hill-climbing algorithm and evolutionary gradient search method in single-objective optimization. Secondly, we use the fitness probability strategy to select individuals for local search in the early stage of the algorithm. Finally, the multi-objective evolutionary algorithm is used to ensure the distribution of nondominated sets after the number of individuals reaches the population. Compared with the NSGA-II and EGS-NSGA-II hybrid algorithms, the results of the ZDT series test functions show that the proposed algorithm has better global and convergence speed.