基于多元统计分析的过程监测研究
文献类型：学位论文
| 作者 | 王中伟 |
| 学位类别 | 硕士 |
| 答辩日期 | 2017-05-24 |
| 授予单位 | 中国科学院沈阳自动化研究所 |
| 授予地点 | 沈阳 |
| 导师 | 宋宏 |
| 关键词 | 主元分析方法 支持向量数据描述 谱聚类 多模态过程监测 局部-全局模型 |
| 其他题名 | Research on Process Monitoring Based on Multivariate Statistical Analysis |
| 学位专业 | 控制工程 |
| 中文摘要 | 现代工业为了应对激烈的市场竞争和保证生产的安全运行，对过程监测和故障诊断越来越重视。由于系统的庞大和过程的复杂性，其也面临着许多难题。如何提高故障检测的准确率降低漏报率一直是该领域的研究热点。主元分析方法（PCA）是经典的多元统计分析方法，被广泛应用于工业生产过程监测。本文针对主元分析方法的一些局限和约束问题，进行了以下几方面的研究。 主元分析方法假设数据服从高斯分布，且建模时所使用的协方差矩阵仅能评估变量间的线性相关关系，无法衡量变量间非线性依赖程度。基于此提出了一种基于最大信息系数(MIC)的PCA的过程监测方法。采用可以度量变量间的非线性相关性的MIC矩阵替换协方差矩阵，从而改善主元分析方法对非线性过程的监测效果。最后将该方法用于田纳西-伊斯曼过程的数据集进行验证仿真，结果表明该方法是可行的和有效的。 支持向量数据描述（SVDD）是一种单值分类方法，对数据无正态分布假设，在非高斯、非线性数据处理中表现出极大的优势。由于主元分析模型局限于线性、高斯分布数据，而实际复杂工业数据往往是非线性和非高斯分布的，基于此提出了一种基于MIC-PCA和SVDD相融合的过程监测方法。首先采用MIC矩阵替换协方差矩阵提高对非线性信息的利用，然后在主元子空间和残差子空间分别建立SVDD模型，改善对非高斯过程的监测效果。最后通过田纳西-伊斯曼过程数据对该方法进行验证仿真，结果表明该方法提高了监测的准确性和实时性。 实际生产过程往往因为生产策略变更、操作条件改变而呈现出多模态特性。经典多元统计分析方法往往假设系统是在单一模态下运行的，而且传统的多模态过程监测方法往往只利用样本数据的局部信息进行建模，忽略了过程操作的全局关联性。基于此，考虑到全局空间的关联性和不同局部空间的差异性，提出了基于局部-全局的MIC-PCA-SVDD的过程监测方法。首先，提出了一种基于谱聚类的模态划分方法，该方法将过程数据划分为多个模态，其后利用主元子空间和残差子空间的物理意义的差异建立局部-全局的MIC-PCA-SVDD的监测模型，在不同的子空间建立不同的统计量和置信限。该方法既描述了变量空间的局部结构信息，包纳了它的全局结构信息。最后通过了典型的多模态过程—青霉素发酵过程的进行试验结果的验证，结果表明该方法在多模态过程监测的有效性和优越性。 |
| 英文摘要 | In order to ensure process safety and stability, process monitoring and fault diagnosis is playing an increasingly important role in modern industry. Due to the complexity of the system and process, it also faces many difficulties。How to improve the correct rate of fault detection to reduce the false negative rate has been the research hotspot in this field. Principal component analysis (PCA) is a classical multivariate statistical analysis method that is widely used in industrial production process monitoring. In view of some limitations and constraints of the principal component analysis model, some researches are carried out about it in the following aspects in this paper. The principal component analysis method assumes that the data obeys the Gaussian distribution, and the covariance matrix used in modeling can only evaluate the linear correlation between variables and can't measure the degree of nonlinear dependence between variables. To solve this shortcoming, a novel process monitoring method based on maximal information coefficient (MIC) and PCA is proposed. The covariance matrix can be replaced by the maximum information coefficient matrix which can measure the nonlinear correlation between variables, so as to improve the monitoring effect of the principal component analysis method on the nonlinear process. Finally, the feasibility and effectiveness of the proposed method are verified by the Tennessee Eastman (TE) process simulation. Support vector data description (SVDD) is a kind of one-class classification method, which has great advantage in non-Gaussian and nonlinear data processing. Complex industrial processes are often non-linear and non-Gaussian, while the traditional principal component analysis method assumes that the data are Gaussian and linear. In this paper, a process monitoring method based on MIC-PCA and SVDD is proposed. First, the covariance matrix is replaced by the MIC matrix which can measure the non-linear correlation between the variables. Then the SVDD models are built in the principal component subspace (PCS) and the residual subspace (RS) to improve the monitoring of non-linear and non-Gaussian processes. Finally, the method is verified by Tennessee-Eastman process data. The results show that the method improves the accuracy and real-time performance of the monitoring. Due to changes in production strategies and operating conditions, the actual production process always comes with multi-modal characteristics. The classical multivariate statistical analysis method often assumes that the system is running in a single mode, and the traditional multimodal process monitoring method often uses only the local information of the sample data to model, ignoring the global correlation of the process operation. Based on this, a process monitoring method based on local-global MIC-PCA-SVDD is proposed considering the correlation of global space and the difference of different local spaces. Firstly, a modal partitioning method based on spectral clustering is proposed. The method divides the multimodal process data into multiple modalities, then with the difference between the principal subspace and the residual subspace to establish the local-global MIC -PCA-SVDD monitoring model and compute different statistics and confidence limits in different subspaces. This method not only describes the local structure information of variable space, but also contains its global structure information. Finally, the results of the method is verified by the typical multimodal process-penicillin fermentation process. The results show that the method is more effective and superior in multimodal process monitoring. |
| 语种 | 中文 |
| 产权排序 | 1 |
| 源URL | [http://ir.sia.cn/handle/173321/20523]  |
| 专题 | 沈阳自动化研究所_数字工厂研究室 |
| 推荐引用方式 GB/T 7714 | 王中伟. 基于多元统计分析的过程监测研究[D]. 沈阳. 中国科学院沈阳自动化研究所. 2017. |
入库方式： OAI收割
来源：沈阳自动化研究所
浏览0
下载0
收藏0
其他版本
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。