流程工业过程数据的多粒度分析方法与应用研究
文献类型：学位论文
| 作者 | 杨田 |
| 学位类别 | 博士 |
| 答辩日期 | 2009-06-07 |
| 授予单位 | 中国科学院软件研究所 |
| 授予地点 | 软件研究所 |
| 关键词 | 数据分析 过程数据 粒度计算 多粒度 数据挖掘 |
| 其他题名 | Research on Multi-Granular Process Data Analysis Method and its Applications in Process Industry |
| 中文摘要 | 现代流程工业产生了大量与生产状态相关的过程数据，其庞大的规模凸显了传统数据分析方法运算复杂度过高、分析规模受限的不足。同时，流程工业生产过程中所包含的一些复杂性，如传感器数据中的高噪音以及异步采样所导致的数据时标不统一等，都增加了过程数据分析的难度。 为了应对上述挑战，本文将粒度计算的思想引入流程工业中的过程数据分析领域。粒度计算是信息处理领域中一种新兴的概念和计算范式，其核心思想是通过适度放松精度要求，在较大粒度上进行分析以提高方法整体的效率和鲁棒性。 本文重点研究了面向过程数据的粒度化分析方法，并围绕该方法的关键性质展开了深入细致的论证，在此基础上给出了选择粒度的方法和准则，最后我们还通过具体实例对上述方法的有效性进行了应用验证。 论文的主要贡献体现在： (1)将粒度计算的思想和过程数据分析方法相结合，提出了一种粒度化的过程数据分析方法，并证明了在特定条件下粒度化的数据分析能够得到原问题解的下界或下确界，为粒度化过程数据分析方法奠定了坚实的理论基础。 (2)深入探讨了不同粒度下过程数据分析的计算复杂性变化情况，提出了一种利用多粒度分析挑选最佳分析粒度的方法。该方法针对现有数据分析方法难于应对海量数据的缺点，可以根据具体应用的精度需求灵活地调整分析粒度，从而降低数据分析的总体计算复杂性。 (3)针对过程数据分析中面临的稳态周期评估问题，提出了一种基于多粒度分析的过程数据稳态周期检测算法。该算法基于我们提出的过程数据质量评估指标，能够灵活地调整查询周期，提供符合应用系统质量需求的数据，同时尽量减少运行开销。 (4)针对大规模软测量面临的计算效率问题，提出了一种基于多粒度分析的软测量方法。该方法通过构建一组粒度化的软测量模型，并依次在多个粒度中求解，最终确定一个既能满足特定精度需求又能确保运算效率最高的粒度。该方法已经得到实践检验并取得良好的效果。 |
| 英文摘要 | Modern process industry has produced tremendous data which are associated with industrial manufacture status and need to be further analyzed. Dataset's large size protrudes the scalability shortcoming of existed analysis methods with high computing complexity. Besides, some other complex factors in manufacture process, such as high noise in sensor data and variant data timestamp caused by asynchronous sampling etc, could also increase the difficulty of process data analysis. To deal with these challenges, this dissertation introduces the idea of granular computing into process data analysis. Granular computing is an emerging concept and computing paradigm. With the help of this new technique, our approach could effectively enhance the efficient and robustness of data analysis by escalating the granularity of analysis and relaxing the requirement on accuracy moderately. This dissertation mainly studies on the approach of granular process data analysis, and then discusses its key properties in detail. On this basis, we present a general approach and related criteria to select the granularity of analysis. Finally, we verify the effectiveness of our approach with some real world applications. The main contributions of this dissertation are listed below. (1) This dissertation links granular computing to process data analysis, and presents a granular-based data analysis approach. Then it proves the method could reach a lower bound (sometimes infimum) of the original problem in some special cases and lays a solid theory foundation for applying granular computing theory into the data analysis. (2) This dissertation thoroughly discusses the variation of computational complexity with different granularity in process data analysis. On this basis, we present a method by which the most appropriate granularity could be selected with the help of multi-granular analysis method and a corresponding application framework. Aiming at existed data analysis methods' shortcoming on handling tremendous data, our method could flexibly adjust the granularity according to the accuracy requirement of applications, in order to reduce the running time of whole data analysis. (3) Aiming at process data quality skyline, this dissertation presents a multi-granular based algorithm to detect this skyline.This algorithm is based on a process data quality assessment we proposed, and could flexibly adjust the skyline to provide data which satisfies the quality requirement of applications with much lower running time. (4) To improve performance of large-scale soft sensing, this dissertation presents a soft sensing approach based on multi-granular data analysis. By building a set of granular soft sensing models and then resolving them in turn, this approach can find an appropriate granularity of soft sensing that can satisfies specific accuracy requirement and has the best performance. This approach has been applied in real manufacture and gets a good effect. |
| 语种 | 中文 |
| 公开日期 | 2011-03-17 |
| 页码 | 131 |
| 源URL | [http://124.16.136.157/handle/311060/6812]  |
| 专题 | 软件研究所_人机交互技术与智能信息处理实验室_学位论文 |
| 推荐引用方式 GB/T 7714 | 杨田. 流程工业过程数据的多粒度分析方法与应用研究[D]. 软件研究所. 中国科学院软件研究所. 2009. |
入库方式： OAI收割
来源：软件研究所
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。