基于特征融合的动态过程质量异常模式识别方法研究

发布时间：2018-05-08 01:25

本文选题：动态过程 + 特征融合　；参考：《郑州大学》2017年硕士论文

【摘要】：随着生产制造过程的日益自动化、连续化与复杂化,动态过程的质量异常模式识别与质量诊断引起了众多学者的关注。质量异常模式识别的准确率主要取决于模式分类特征与分类器这两个关键因素,从质量模式原始数据中所提取出的特征分量,不仅能够有效地反映出质量模式的属性与状态,增强不同模式之间的区分度,还能够在很大程度上减少数据的冗余性与繁杂性,已经成为提升质量模式识别精度的有效手段之一。然而,由于动态过程大数据的复杂性,任何一类异常模式,仅依靠单一类型的数据特征难以获得较高的识别精度。因此,如何针对动态数据流提取低维数且细节信息较强的特征数据,采用融合特征的方法以提高动态过程异常模式识别效率是亟待解决的问题。本文在收集整理大量国内外研究文献的基础上,以模式识别和质量诊断为理论依据,将系统地研究基于特征融合的动态过程质量异常模式识别方法。首先,在国内外动态过程模式识别、特征提取方法与质量诊断研究综述的基础上,界定了动态过程的质量异常模式;随后,提出了基于特征融合的动态过程质量异常模式识别方法,并采用粒子群优化算法寻找支持向量机最佳参数组合。最后,通过仿真实验验证了本文所提方法的有效性。研究结果表明:(1)本文所提出的基于特征融合优化与多支持向量机的质量异常模式识别方法比传统识别方法的效率更高;(2)通过粒子群优化算法寻找支持向量机的最优参数组合,并将融合约简特征作为多支持向量机的输入向量能够获得比其他识别模型更好的识别效果;(3)该模型利用粗糙集对串联特征组合进行约简,能够降低特征的维数,并剔除冗余与无关特征,提升整个模型的识别精度。本文的研究特色与创新之处在于:(1)提出了基于融合特征的动态过程质量异常模式识别方法;(2)将粗糙集约简方法应用于质量异常模式识别的特征组合融合优化,消除对分类贡献较小或是包含冗余信息的特征,进而得到便于进行分类的特征集合;(3)构建了多支持向量机作为质量异常模式识别的分类器,并利用粒子群算法寻找其最优参数组合,实现了动态过程质量异常模式的识别。本文的研究克服了单一类型的数据特征仅用部分信息来反映动态过程运行状态的缺陷,并有效地压缩了冗余信息,实现了实时质量异常数据的处理与诊断,为石油、化工、烟草等自动化行业提供实时质量监控与故障诊断技术。
[Abstract]:With the increasing automation, continuity and complexity of the manufacturing process, many scholars have paid close attention to the quality anomaly pattern recognition and quality diagnosis of the dynamic process. The accuracy of quality anomaly pattern recognition mainly depends on the two key factors of pattern classification feature and classifier. The feature components extracted from the quality pattern raw data can not only effectively reflect the attribute and state of the quality pattern. It has become one of the effective methods to improve the accuracy of pattern recognition to enhance the degree of discrimination between different patterns and to reduce the redundancy and complexity of the data to a great extent. However, due to the complexity of the dynamic process big data, it is difficult to obtain high recognition accuracy for any kind of abnormal pattern only depending on a single type of data features. Therefore, how to extract feature data with low dimension and strong detail information for dynamic data flow, and how to improve the efficiency of dynamic process anomaly pattern recognition by using the method of fusion feature is an urgent problem to be solved. On the basis of collecting and sorting out a large number of domestic and foreign literatures, based on the theory of pattern recognition and quality diagnosis, this paper will systematically study the dynamic process quality anomaly pattern recognition method based on feature fusion. Firstly, on the basis of the summarization of research on dynamic process pattern recognition, feature extraction and quality diagnosis at home and abroad, the quality anomaly pattern of dynamic process is defined. Based on feature fusion, a dynamic process quality anomaly pattern recognition method is proposed, and particle swarm optimization algorithm is used to find the best parameter combination of support vector machine. Finally, the effectiveness of the proposed method is verified by simulation experiments. The research results show that the proposed quality anomaly pattern recognition method based on feature fusion optimization and multi-support vector machine is more efficient than the traditional recognition method. The particle swarm optimization algorithm is used to find the optimal parameter combination of support vector machine. The fusion reduction feature can be used as input vector of multi-support vector machine to obtain better recognition effect than other recognition models. The model can reduce the dimension of feature by using rough set to reduce series feature combination. The redundant and irrelevant features are eliminated and the recognition accuracy of the whole model is improved. In this paper, the research features and innovations lie in: (1) A method of dynamic process quality anomaly pattern recognition based on fusion feature is proposed. The rough set reduction method is applied to the feature combination fusion optimization of quality anomaly pattern recognition. To eliminate the features which contribute little to the classification or contain redundant information, and then get the feature set which is convenient for classification, a multi-support vector machine (SVM) is constructed as the classifier for quality anomaly pattern recognition. Particle swarm optimization (PSO) algorithm is used to find the optimal parameter combination to identify the abnormal pattern of dynamic process quality. The research in this paper overcomes the defect that the single type of data features only use partial information to reflect the running state of dynamic process, and effectively compresses the redundant information, and realizes the processing and diagnosis of real-time abnormal quality data. Tobacco and other automation industries provide real-time quality monitoring and fault diagnosis technology.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F273.2

【相似文献】