Abstract: As for the low accuracy and poor stability of pipeline leak diagnosis, this paper proposes a pipeline leak detection method combining SVM with DS evidence theory. Five kinds of characteristic parameters are extracted by means of wavelet analysis, and then the extracted characteristic parameters are input into SVM to train and get the primary classification results to construct the Basic Probability Assignment（BPA）, and the rule of combination is used to fuse the data in the different time and space fields, the final decision results are got according to the judgment threshold. The proposed method is applied to pipeline leak detection system under lab conditions, and the results show that the accuracy rate of pipeline leak diagnosis reaches 96.42%, the proposed method has higher recognition rate and stability compared with other pattern recognition methods.

Key words: leak diagnosis, support vector machine, evidence theory, pattern recognition, judgment threshold

摘要： 针对管道泄漏诊断准确率低和稳定性差的问题，提出一种结合SVM与DS证据理论的管道泄漏识别方法。运用小波分析法提取5种反映管道运行状态的特征参数，输入SVM分类器中进行初次分类，然后根据识别结果构造基本概率指派（BPA），再运用组合规则对不同时空域数据进行DS融合，最后根据判决门限得出决策结果，将这种SVM-DS诊断方法应用于实验室管道泄漏检测系统中。结果表明，SVM-DS的管道泄漏识别准确率达到96.42%，与其他模式识别方法相比，该方法具有更高的识别率和更强的稳定性。

关键词: 泄漏诊断, 支持向量机, 证据理论, 模式识别, 判决门限