基于适应性动态步长的变异果蝇优化算法

Mutation fruit fly optimization algorithm based on adaptive dynamic step size

针对基本果蝇优化算法(FOA)容易陷入局部最优值、后期收敛速度变慢和收敛精度较低的缺点,提出了一种基于适应性动态步长的变异果蝇优化算法(MFOAADS)。首先,利用佳点集法选取种群初始位置,降低算法初始点选取的随机性和陷入局部最优值的概率;然后,采用适应性动态步长优化策略,提高收敛速度和求解精度;最后,若算法陷入了早熟,则对种群最优个体按一定概率执行柯西变异扰动,赋予其跳出局部最优的能力。经5个经典函数测试表明,固定迭代次数时MFOAADS的收敛精度与收敛速度明显优于FOA;固定目标精度时,MFOAADS相对于FOA平均迭代次数有着大幅下降且成功率达97%以上。实验结果表明,所提算法求解精度、运行效率以及可靠性相对于基本FOA算法都有着显著提高。

The basic Fruit Fly Optimization Algorithm（ FOA） has the shortcomings of being easy to fall into local optimal value, slow convergence and low convergence accuracy. Aiming at the problems, a Mutation Fruit Fly Optimization Algorithm Based on Adaptive Dynamic Step Size（ MFOAADS） was proposed. Firstly, the selection of the initial position of the population was improved using the optimal point set method, which reduced the randomness of initial point selection and the probability of getting trapped in local optimal value. Secondly, the adaptive dynamic step size optimization strategy was adopted to improve the convergence rate and the accuracy of the solution. Finally, if the algorithm fell into premature convergence, the Cauchy mutation perturbation would be utilized in a certain probability for the sake of making them jump out of local optimum. The test results of the five classical functions showed that convergence accuracy and convergence speed of MFOAADS were obviously superior to the FOA when the number of iterations was fixed. And in the comparison experiments with FOA, the average number of iterations of MFOAADS decreased significantly with a success rate of more than 97% when target accuracy was fixed. The experimental results show that, compared with the basic FOA, the proposed algorithm significantly improves the accuracy, efficiency and reliability.