微加工超声传感器的工作机制和理论模型的研究

Nowadays, Data mining (DM) and Knowledge Discovery in Database (KDD) are two of the most active directions in some fields such as artificial intelligence and database and will be.used widely in the near future. They are also the overlapping parts of some research fields such as artificial intelligence, statistics, machine learning and database.Time Series Data Mining(TSDM) is one of important Data Mining research . Its topics include time series similarity search, clustering, classification, association, event detection, outlier.discovery, prediction etc.Time series are complex types of data. They often have high dimensionality,noise and various distortions such as offset translation,amplitude scaling, stretching or compressing in the.time-axis,linear drift and discontinuities. These are difficult factors for data mining on time.series. In this dissertation, problems related to TSDM based on wavelet analysis, especially time series similarity measure and similarity search algorithms , are studied in detail.TSDM techniques based on wavelets analysis are proposed. First, time series dimensionality reduction methods by wavelet and similarity measure method based on Euclid distance are discussed. Second, a multi-resolution and multi-level time series similarity search technique based on wavelet transformations and edit distance is proposed. Third, a novel similarity measure model is presented. In the model, time series extremum points are considered as critical points and features of the sequence. These points are the.approximation representation of raw data. They can be transformed into a binary tree. Two time series are simliar if the two binary trees, which result from them by this method, are.identical. Multi-resolution clustering and similarity search tenique are also presented based on the model. Finally the method of aberrant event detection based on wavelet transform modulus.maxima are studied.A model Vis-TSMiner based on wavelet analysis for visual data mining on time series is proposed. The model consists of five components: original data visualization, visaul.data preprocess, visual data reduction, visual pattern discovery and pattern visualization. By wavelets the model performs hierarchical representation of time series dataset for visualization, data reduction and multi-scale pattern discovery. This model can help users.view the high dimensional data , unstand the intermediate results, and interpret the discovered patterns.

近年发展的微加工超声传感器以其优越性能在机器人，无损检测，材料研究及生物医学等领域展现出广阔的应用前景，但其工作机制未搞清楚，以致它的研究停留在经验阶段。本项目拟在英国合作研究和省基金项目研究的基础上深入研究它的动力学机制和机电耦合机理，建立可靠的理论模型，从而为这种传感器的研究及其应用提供理论的基础。