Abstract: In order to classify the time-series objects，a binary-tree SVM with its advantages for multiclassification is presented.Since SVM does not make use of the time-series property of the classification and leads to the descent of recognition accuracy，by means of the data windows the label of an object，which is classified by SVM，is corrected by the distribution of the time-series objects in data window.Then the accuracy of time-series object recognition is greatly improved.This approach is applied in stratum recognition with well-logging data successfully.It is proved efficient that the recognition accuracy of stratum is improved greatly with the two steps of “recognition” and “correction”.

Key words: time-series data, support vector machine, data windows, well-logging interpretation

摘要： 通过建立多分类二叉树的SVM识别模型，充分利用SVM的优点，实现具有时序特征对象的识别。由于SVM没能利用时序特性有用的分类信息，导致部分类别判断的失误，因此通过引入数据窗口，利用时序对象的类别分布来校正SVM识别的奇异对象，进一步提高时序对象识别的精度。利用测井数据，以地层识别为应用背景，通过“识别”和“校正”两个阶段，大大提高地层的识别精度，取得很好的应用效果。

关键词: 时序数据, 支持向量机, 数据窗口, 测井解释