时间序列模式匹配技术分析-软件工程专业论文.docx

准确度，同时有效地提高了计算效率。 在深入分析动态序列、动态匹配和匹配算法特点的基础上，提出了模式序列的模式 阈值的概念，设计和实现了适合于动态时间序列的基于模式阈值的定位算法，并对该算 法进行了优化。 以上述研究成果为基础，设计和实现了一个同时支持子序列匹配、全序列匹配和动 态匹配的时间序列模式匹配原型系统 TSPM(Time Series Pattern Matching),为评估各 类匹配方法的实际效果提供了实验研究平台。 关键词：时间序列，模式匹配，子序列匹配，全序列匹配，子序列定位，动态匹配，提 前终止技术 AbstractTime series pattern matching is an important and fundamental problem in time series data mining research. It is the basic technique and method used in clustering, classification, rule discovery, abnormal detection, knowledge discovery and other data mining research tasks. Time series pattern matching is still an active research topic with existing research results showing potential for widespread applications.The GEMINI framework proposed by Faloutsos et al has become a mainstream framework in this research area. This dissertation models time series with GEMINI framework and related research is carried out based on this framework. To carry out the classification, sub-series localization, distance measure, the distance measure optimization and dynamic matching method is important to study.In this dissertation, we categorize time series pattern matching into sub-series matching and whole-series matching. Sub-series matching is a process to search for a sub-series identical or similar to a given series from a longer main series. The key to the solution is to reduce the number of matching comparisons; the most important operation in sub-series matching is to localizing the sub-series. Whole-series matching is a process to search for a series identical or similar to a given series in a database of time series. Whole-series matching is usually converted to distance measure problem and its key is to enhance the accuracy and efficiency of estimating distance measure between time series.Localizing sub-series is the most important operation in sub-series matching which has wide spread application. We propose a new greedy algorithm for sub-series localization based on classic greedy algorithm. Our greedy algori