基于麻雀搜索算法的概率神经网络优化方法及其应用  

Probabilistic Neural Network Optimization Method and Its Application Based on Sparrow Search Algorithm

作　　者：林羽杰 颜勇君 童炼 LIN Yujie;YAN Yongjun;TONG Lian(School of Computer,Hunan University of Technology,Zhuzhou Hunan 412007,China;School of Computer Science and Engineering,Changsha University,Changsha Hunan 410022,China)

机构地区：[1]湖南工业大学计算机学院,湖南株洲412007 [2]长沙学院计算机科学与工程学院,湖南长沙410022

出　　处：《长沙大学学报》2024年第5期19-25,共7页Journal of Changsha University

基　　金：湖南省社科基金教育学专项课题“大数据背景下的留学生跨文化适应评价体系构建与分类方法研究”(JJ194000)。

摘　　要：现如今,模式识别已被广泛应用于语音识别、机械诊断、遥感和医学诊断等领域。概率神经网络(Probabilistic Neural Network,PNN)作为一种常用的模式识别工具,已被应用于各类工程模式识别,且取得了一定效果。然而,传统PNN模型一般采用默认的平滑因子参数,这容易导致识别率低和分类能力不足的问题。为了解决上述问题,提出了一种基于麻雀搜索算法(Sparrow Search Algorithm,SSA)优化概率神经网络的模式识别方法(SSA-PNN)。该方法通过引入麻雀搜索算法,可以在搜索空间内高效地搜寻到最优的平滑因子,并进行PNN模式识别。将该方法应用于模拟信号和真实滚动轴承振动信号数据,实验结果表明,相比于传统PNN模型和遗传算法优化PNN方法(GA-PNN),改进后的SSA-PNN方法识别率整体提升了6.6%和2.2%,达到了93.3%的高识别率,实现了更好的分类效果。Recently,pattern recognition has been widely applied in fields such as speech recognition,mechanical diagnosis,remote sensing,and medical diagnosis.Probabilistic Neural Network(PNN),as a commonly used tool,has been applied to various engineering pattern recognition and achieved certain results.However,traditional PNN models normally use default smoothing factor parameters,which can easily lead to low recognition rate and insufficient classification ability.To address the aforementioned problems,the paper proposes a Sparrow Search Algorithm(SSA)optimized Probabilistic Neural Network pattern recognition method(SSA-PNN).This method can efficiently search for the optimal smoothing factor in the search space and perform PNN pattern recognition.To verify the proposed method,the experiments based on simulated signal and real rolling bearing signal were carried out.The results show that compared to traditional PNN models and genetic algorithm optimized PNN methods(GA-PNN),the proposed SSA-PNN method improves overall by 6.6%and 2.2%,achieving a high recognition rate of 93.3%,and thus achieving better classification performance.

关 键 词：模式识别 概率神经网络 麻雀搜索算法 平滑因子 贝叶斯决策 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]