从振动数据中发现目标的振动特性，有利于识别监控环境中的目标和状态，在安全防御方面有着重要的学术价值和现实意义。本文将数据挖掘技术应用于振动数据的模式匹配研究，在研究和比较几种典型分类算法的基础上，探索了振动数据模式匹配的新方法，并对振动源的振幅能量和距离建立模型，判断振动源的具体位置，同时开发了相应的软件原型系统。主要研究内容如下： 针对振动源模式识别率低的问题，提出了基于投影比例的K最近邻分类方法(KNNPS)。该方法引入了投影比例的概念，首先使用独立分量分析方法对信号进行噪声分离，然后利用连续属性离散化方法进行降维，最后通过投影比例度量信号间的相似度，得到振动信号的模式类别。利用正确率、错误率、检测率和模式匹配率来评价KNNPS、K最近邻(KNN)和朴素贝叶斯算法的性能。实验结果表明，KNNPS算法的性能要优于KNN和朴素贝叶斯算法。对分类结果进行分析，可以根据正确分类的类别个数得出振动源的数目，并根据同一类别下振动幅值的强弱来判断振动数据的距离属性。 针对利用振动幅值的强弱无法正确估计振动源位置的问题，建立了振动源振幅能量的衰减模型函数，得到振幅能量和距离的关系。对于传感器采集的振动数据，利用函数模型来估计振动源的位置，实现了利用传感器对振动源距离的有效估计，并通过多传感器对振动源的坐标进行定位。 最后，基于上述研究结果，将独立分量分析方法和分类方法应用于振动数据模式匹配系统中。系统模块包括信号分离模块，振动源比较模块，信号匹配模块。测试结果表明系统运行正确，达到了预期目标。

Discovering the characteristic of vibration data are beneficial to identify the target and state in monitor environment, which have great academic value and practical significance in the area of security defense. With careful study and comparison of several classification methods, this paper explores a new method to identify pattern of vibration data. The model of vibration amplitude and distance is built, and the location of the vibration source is judged by the model. At the same time, a prototype software system was developed. The main contents are as follows: According to the problem of the low pattern recognition rate of the vibration source, a k-Nearest Neighbors method based on projection scale (KNNPS) is proposed. The method introduces the concept of projection scale. First, the independent component analysis (ICA) is used to separate noise from the signal. Then the discretization of continuous attributes is used to descend dimension. Finally, the similarity is measured by projection scale, and the pattern of vibration signal is matched. The accuracy rate, error rate, detection rate and pattern matching rate are used to evaluate the performance of KNNPS, KNN, and NB algorithms. The experimental results show that the KNNPS algorithm outperforms the KNN and NB algorithm. Through analyzing the results of the classification，the number of the vibration source can got by the category of classification, and the distance of vibration source can got by strength of the vibration amplitude of the same category. The location of the vibration source cannot be correctly estimated by strong and weak of the vibration amplitude, so the amplitude of fade model is built to obtain the distance of vibration source. The distance of vibration source is estimated effectively by single sensor, and the coordinate of vibration is located by multi-sensor. Based on these researches above, the independent component analysis method (ICA) and classification method are applied in the vibration data pattern matching system. The system consists of the signal separation module, the vibration source judgment module, and signal matching module. The experimental results show that this system works well and could reaches the expected target.