基于二次优化的随机优化算法结果的改进

统计研究发现，随机优化算法多次运行后的优化结果满足正态分布，且期望值更接近最优解。为此，提出一种基于统计学理论并结合牛顿法的二次优化方法来改进随机优化算法的求解结果，以克服将多次优化结果的平均值作为最优解时不能满足精度要求的缺陷。以遗传算法对4个经典测试函数的多次优化为例，分别运用平均法和二次优化法来综合其优化结果。多次实验表明，二次优化法在处理多次随机运行结果时，比平均法精度更高、稳定性更好。

Improving results of stochastic optimization algorithms via secondary optimization

Through analyzing the stochastic optimization results, it is been found that optimization results satisfy normal distribution, and the expected value is the optimal solution. A secondary optimization method based on the statistical theory and the Newton method is proposed to improve the optimization results of stochastic optimization algorithms, which can overcome the average method’s shortcoming that the precision requirements are often can not be met. Taking multiple optimization results of four classic test functions optimized by genetic algorithm as examples, the average method and the secondary optimization method are respectively used to synthesize the optimization results. Experiments show that, in dealing with multiple stochastic optimization results, the secondary optimization method has higher accuracy and better stability than those of the average method.