多空间协同进化的自然计算方法

Natural Computing Method of Multispace Coevolution

在自然计算方法中,种群规模大,计算复杂度高;种群规模小,容易陷入局部最优.本文提出多空间协同进化(Multispace Coevolution,简称MSC)的自然计算方法,该方法适用于各种基于种群进化的优化算法,不依赖于算法进化的具体步骤,具有普适性.在传统的生物种群进化的基础上,将大种群分解为个数有限的小种群,部分小种群组成进化空间,另一部分构成指导空间,两个空间拥有不同的功能,指导空间通过特定的信息传递方式将经验概括信息传递到进化空间,从而使整个种群协同进化.将该策略分别应用到粒子群优化算法(PSO)和遗传算法(GA)中,并与标准粒子群算法、遗传算法以及目前主流的针对大规模问题进行优化的7个算法对比,在高维测试函数中,结果表明,寻优性能方面新的种群进化算法相比其他算法提高80%左右,具有普适性.

In the natural calculation method, the population size is large and the calculation complexity is high;the population size is small and easy to fall into local optimum.In this paper, a natural computing method of multispace coevolution(MSC)is proposed.This method is suitable for all kinds of optimization algorithms based on population evolution, does not depend on the specific steps of algorithm evolution, and has universality.On the basis of the traditional evolution of biological population, the large population is decomposed into a limited number of small populations, some of which constitute the evolution space, and the other one constitutes the guidance space.The two spaces have different functions.The guidance space transmits the general information of experience to the evolution space through a specific way of information transmission, so that the whole population can co evolve.This strategy is applied to particle swarm optimization(PSO)and genetic algorithm(GA)respectively, and compared with standard particle swarm optimization(PSO),genetic algorithm(GA)and the current mainstream seven algorithms for large-scale optimization.In the high-dimensional function, the results show that the new evolutionary algorithm for population optimization has increased by 80% and universality compared with other algorithms.