【摘要】 针对供水管道漏损现场检测误判、耗时、低效等问题，基于地面振动信号和时频卷积神经网络设计了漏损信号自动识别模型。将采集的地面振动信号进行连续小波变换，得到了包含漏损特征的时频图像，将其输入到卷积神经网络，并对网络超参数进行优化，最终得到了漏损识别模型。结果表明，最终模型在测试集的平均准确率为97.3%，针对检漏人员难以区分的漏损点与漏损点附近的可疑信号平均识别率分别为91.0%和92.3%，具备良好的诊断漏损能力。相比支持向量机、决策树等方法，所提出的方法具有更高的准确率。更多还原

【Abstract】 This paper designed an automatic identification model of leakage signal based on ground vibration signal and time-frequency convolutional neural network to solve the problems of misjudgment, time consuming and low efficiency of water supply pipeline leakage detection. The timefrequency image containing leakage characteristics was obtained by continuous wavelet transform of the collected ground vibration signals, and was input into the convolutional neural network to optimize the network hyperparameters, and the leakage identification model was eventually obtained. The average accuracy of the final model in the test set was 97.3%, and the average recognition rates of the leak point difficult to distinguish by detectors and the suspicious signal near the leak point were 91.0% and 92.3%,respectively, indicating that the model had a good ability of leak diagnosis. Compared with support vector machine, decision tree and other methods, the proposed method had higher accuracy.更多还原