摘要
针对传统多分类相关向量机(relevance vector machine,RVM)采用"最大票数赢(MVW)"决策策略的不足,为了提升相关向量机的多分类能力,首先改进了RVM的多分类决策策略,并利用具有Lévy飞行特征的果蝇算法(LFOA)对RVM核参数进行寻优,建立了LFOA-RVM分类模型。在适应度函数的评判下,果蝇种群经过多次迭代对指定范围内的核参数进行全局搜索寻优,完成模型建立。四组UCI标准数据集的MATLAB仿真实验结果表明,改进后的多分类决策策略和优化方法有效、可靠,能够提升RVM的分类能力;进一步将此模型应用于液压泵故障诊断,同样取得了较好的分类效果,验证了分类模型的有效性。
Aiming at the drawbacks of the traditional multi-classification of relevance vector machine using the decision strategy of "maximum number of votes wins"and in order to improve the multi-classification performance of relevance vector machine,this paper improved the multi-classification decision strategy of RVM,and optimized the RVM kernel parameter by LFOA algorithm to establish the classification model. By the evaluation of fitness function,the fruit fly group began a global search for the kernel parameter in the specified range through several iterations. The MATLAB simulation experiment through four UCI standard datasets verifies the effectiveness of the classification model,and the multi-classification decision strategy and proposed optimization method can improve the performance of RVM. Further,it applied this model to the fault diagnosis of hydraulic pump,and achieved a good classification result,which verifies the validity of the proposed classification model.
出处
《计算机应用研究》
CSCD
北大核心
2017年第12期3721-3724,3734,共5页
Application Research of Computers
基金
河北省自然科学基金资助项目(E2016506003)
关键词
相关向量机
决策策略
核参数优化
分类模型
RVM
decision strategy
kernel parameter optimization
classification model
