提出了一种结合C-均值聚类的自适应共振(Adaptive Resonance Theory 2,ART2)神经网络无监督学习分类方法,用于风电机组齿轮箱设备群的故障诊断.利用某风电场齿轮箱运行数据,采用ART2神经网络对样本数据进行初步分类,再采用C-均值聚类算...提出了一种结合C-均值聚类的自适应共振(Adaptive Resonance Theory 2,ART2)神经网络无监督学习分类方法,用于风电机组齿轮箱设备群的故障诊断.利用某风电场齿轮箱运行数据,采用ART2神经网络对样本数据进行初步分类,再采用C-均值聚类算法对神经网络分类结果进行修正,得到最终诊断结果,并与ART2神经网络分类结果进行了比较.结果表明:所提出的方法解决了原始神经网络算法存在"硬竞争"导致分类精度下降的问题,准确度高于传统的ART2神经网络,可以准确识别出故障齿轮箱.展开更多
The paper describes results obtained in the development of adaptive fuzzy-neural navigation subsystem for mobile legged robot. In order to keep the motion sufficiently smooth, free of sharp turnings and transversal sw...The paper describes results obtained in the development of adaptive fuzzy-neural navigation subsystem for mobile legged robot. In order to keep the motion sufficiently smooth, free of sharp turnings and transversal swings when moving between closely located obstacles, fuzzy rules are updated on-line. To this end, the fuzzy rules are expressed through a layered feed-forward neural network and parameters are updated on line in two steps--the rough and fine updating. That is followed by the description of the learning fault diagnosis using binary neural network based on the Carpenter and Grossbergs＇ adaptive resonance theory.展开更多
文摘提出了一种结合C-均值聚类的自适应共振(Adaptive Resonance Theory 2,ART2)神经网络无监督学习分类方法,用于风电机组齿轮箱设备群的故障诊断.利用某风电场齿轮箱运行数据,采用ART2神经网络对样本数据进行初步分类,再采用C-均值聚类算法对神经网络分类结果进行修正,得到最终诊断结果,并与ART2神经网络分类结果进行了比较.结果表明:所提出的方法解决了原始神经网络算法存在"硬竞争"导致分类精度下降的问题,准确度高于传统的ART2神经网络,可以准确识别出故障齿轮箱.
文摘The paper describes results obtained in the development of adaptive fuzzy-neural navigation subsystem for mobile legged robot. In order to keep the motion sufficiently smooth, free of sharp turnings and transversal swings when moving between closely located obstacles, fuzzy rules are updated on-line. To this end, the fuzzy rules are expressed through a layered feed-forward neural network and parameters are updated on line in two steps--the rough and fine updating. That is followed by the description of the learning fault diagnosis using binary neural network based on the Carpenter and Grossbergs＇ adaptive resonance theory.
文摘天津位于京津冀区域,近年来面临的颗粒物污染问题受到广泛关注,研究其大气环境中颗粒物的化学组成及来源具有重要意义.为明确天津市夏季环境受体中颗粒物的混合状态及可能来源,于2017年7月利用单颗粒气溶胶质谱仪(single particle aerosol mass spectrometer,SPAMS)在津南区采集到成功电离有粒径及完整质谱信息颗粒209887个,利用ART-2a对有质谱数据的颗粒按照质谱特征的相似性进行聚类共获得369个颗粒物类别,随后按照类别的化学组成(质谱谱图)的相似性进行人工合并获得19个颗粒物类别,包括:K-EC(0.20%)、K-EC-Sec(0.18%)、K-NO3-PO3(12.00%)、K-NO3-SiO3(2.98%)、K-Sec(0.16%)、EC(39.60%)、EC-Sec(3.46%)、EC-HM-Sec(3.93%)、HEC(1.49%)、HEC-Sec(1.38%)、OC-Amine-Sec(3.58%)、OC-Sec(0.36%)、OCEC-Sec(0.71%)、Dust-HEC(21.35%)、Dust-Sec(0.72%)、Cl-EC-NO3(1.22%)、Na-Cl-NO3(3.20%)、HM-Sec(2.58%)和PAH-Sec(0.90%)颗粒.得到的各个颗粒类别可归因于气溶胶颗粒的不同来源及不同的传输和反应过程,综合分析采集到的颗粒贡献源主要包括机动车排放源、生物质燃烧源、工业排放源、扬尘源、燃煤源和二次源等.其中K-EC、EC、HEC和Dust-HEC等颗粒主要来自一次源直接排放,K-Sec、OC-Amine-Sec、OC-Sec、OCEC-Sec和Na-Cl-NO3等颗粒大都是一次源排放颗粒经历了不同程度的老化或与二次组分进行了不同程度的混合.
