郑书闽,颜建国,郭鹏程,徐燕,李江,刘振兴.基于VMD及深度学习的供水管道小尺度泄漏检测研究[J].水利学报,2024,55(8):999-1008 |
基于VMD及深度学习的供水管道小尺度泄漏检测研究 |
Small-scale pipeline leak detection based on VMD and deep learning |
投稿时间:2023-10-30 |
DOI:10.13243/j.cnki.slxb.20230676 |
中文关键词: 泄漏检测 小尺度泄漏 变分模态分解 深度学习 供水管道 |
英文关键词: leak detection small-scale leakage variational mode decomposition deep learning water supply pipeline |
基金项目:国家自然科学基金项目(51839010);陕西省创新能力支撑计划项目(2024RS-CXTD-31);陕西高校青年创新团队项目(2020-29);新疆水专项(2020.C-001) |
|
摘要点击次数: 1650 |
全文下载次数: 1885 |
中文摘要: |
针对常压、小尺度泄漏条件下,泄漏信号不明显、难以有效检测的问题,开展了供水管道泄漏检测研究,获取了100~220 kPa压力、40~80 m3/h流量条件下的泄漏实验数据,并分析了压力泄漏信号的变化规律。为减少噪声干扰并增强泄漏信号特征,采用变分模态分解(VMD)对实验数据进行降噪,并进行标准化处理。基于长短期记忆网络(LSTM)、双向长短期记忆网络(BiLSTM)、门控循环单元(GRU)等典型循环神经网络,结合卷积神经网络(CNN),构建了CNN-LSTM、CNN-BiLSTM以及CNN-GRU等3种深度学习泄漏检测模型,并进行了预测性能评价,其中,CNN-GRU模型对全部实验数据的预测精度高达99.56%。结果表明:所构建的泄漏模型能够有效判断常压、小尺度条件下的管道是否发生漏损;利用CNN进行特征提取,能够有效提取泄漏特征,从而提升泄漏检测模型的预测精度和泛化性。研究工作可为管道泄漏检测系统智慧管理提供支撑。 |
英文摘要: |
To address the challenge of detecting leakage signals under normal pressure and small-scale leaks,this paper focuses on the detection of water supply pipeline leaks.The experimental data of leakage under the conditions of 100-220 kPa pressure and 40-80 m3/h volume flow were obtained,and the variations in pressure signals under small-scale leak conditions were analyzed.The experimental data is denoised by using Variational Mode Decomposition(VMD)to reduce noise interference and enhance leak signal characteristics,followed by standardization process.The study combines typical recurrent neural networks,including Long Short-Term Memory(LSTM),Bidirectional Long Short-Term Memory(BiLSTM),and Gated Recurrent Unit(GRU),with Convolutional Neural Network(CNN)to construct three deep〖JP2〗 learning leakage detection models CNN-LSTM,CNN-BiLSTM,and CNN-GRU.These models were evaluated for their predictive performance,among them,the CNN-GRU model exhibited the highest predictive accuracy of 99.56% for all experimental data.The results indicate that the models demonstrate high accuracy in detecting leaks under normal pressure and small-scale leak conditions.CNN proves to be instrumental in extracting pertinent features efficiently and accurately,thereby improving the prediction accuracy of the leakage detection model.The research provides valuable support for the intelligent management of pipeline leakage detection system. |
查看全文
查看/发表评论 下载PDF阅读器 |
关闭 |