Abstract: Basic FOA（Fruit Fly Optimization Algorithm） often derives a local extreme and the basic FOA cannot solve optimization problem when there exists negative number in the domain because smell concentration judge value S is ?non-negative. In order to solve these problems, multiple improved strategies are proposed to improve basic FOA. The calculating formula of smell concentration judge value is modified to solve the problem that S cannot be negative number in basic FOA. In order to solve high-dimensional inter-perturbation, dimension by dimension perturbation strategy is used in update stage. Convergence judge threshold is used in the iteration process. The algorithm derives a local extreme if optimal value is not better after many time iterations. And then, some excellent flies optimize near the optimal fly and other flies find the optimum solution with chaotic disturbance in the solution domain. Because the optimization performance of algorithm is influenced by the convergence judge threshold, Experiments are done with various thresholds and the premium threshold is determined according to experiment results. Through simulation experiment on test functions, the proposed algorithm has higher searching precision and faster convergence speed than basic FOA.

Key words: Fruit Fly Optimization Algorithm（FOA）, parameter modification, dimension by dimension perturbation, chaotic disturbance, convergence judge threshold

摘要： 传统的果蝇优化算法（Fruit Fly Optimization Algorithm，FOA）容易陷入局部最优，而且传统果蝇个体味道浓度判定值S是非负数，不能解决最优解是负数的优化问题。针对以上问题，多重改进策略被应用到果蝇优化算法中。为了解决味道浓度判定值不能是负数的问题，对味道浓度公式进行了修正；为了避免高维函数维间互扰问题，迭代优化的过程中对果蝇个体在最优值附近寻优采取逐维扰动的方法；为了避免陷入局部最优，迭代过程中加入了收敛判断因子，如果多次迭代没有改善，说明陷入了局部最优。此时，一部分果蝇个体继续在最优解附近寻优，另外一部分个体在解空间混沌扰动寻找全局最优解。收敛判断因子阈值的取值会影响优化的速度和精度，通过实验确定了收敛判断阈值。通过对测试函数结果验证表明，改进的果蝇算法比FOA算法具有更高的搜索精度和更快的收敛速度。

关键词: 果蝇优化算法, 参数修正, 逐维扰动, 混沌扰动, 收敛判断因子