多策略混合改进蝴蝶算法的多层感知器训练优化

Multi-layer perceptron training optimization based on multi-strategy hybrid improved butterfly algorithm

传统调参方法容易生成局部最优,提出多策略混合改进BOA算法的MLP训练方法。针对标准BOA算法寻优精度、收敛速度的不足,设计多策略混合改进蝴蝶优化算法MSHBOA。利用佳点集进行种群初始化,避免随机种群的遍历不足;引入量子搜索局部开发和自适应惯性权重全局搜索提升个体寻优能力;设计概率阈值自适应调整机制,动态协调全局搜索与局部开发比重;设计柯西变异和混沌映射的个体混合扰动机制,增强算法脱离局部最优的能力。将MSHBOA应用于多层感知器MLP训练优化,结果验证训练算法在均方误差、分类准确率、局部最优抵抗性和稳定性方面均表现更好。

The traditional parameter adjustment method is easy to lead to local optimization,a MLP training method based on multi-strategy hybrid improved butterfly optimization algorithm was proposed.Aiming at the shortcomings of low optimization accuracy and convergence speed in the basic butterfly optimization algorithm,a multi-strategy hybrid improved butterfly optimization algorithm MSHBOA was proposed.The population was initialized using the good point set to avoid the insufficient ergodic of the random population.The local development of quantum search and the global search of adaptive inertia weight were introduced to improve the individual optimization ability.An adaptive adjustment for probability threshold was designed to dynamically coordinate the proportion of global search and local development.A hybrid individual disturbance mechanism based on Cauchy mutation and chaos map was designed to make the algorithm have the ability to break away from local optimization.MSHBOA was applied to the MLP training optimization of multi-layer perceptron.The results show that the training method has better performances in mean square error,classification accuracy,local optimal resistance and stability.