基于模型迁移学习的铸铁管漏水信号检测研究
Leak Detection of Cast Iron Pipes Based on Model Transfer Learning
为解决管道漏水检测模型泛化能力低、各种材质管道漏水数据难以全面采集导致数据缺失,设计并提出基于一维卷积神经网络构建的PVC管道漏水检测模型(1D_CNN),检测铸铁管是否漏水。为提高PVC漏水检测模型的泛化能力,通过基于模型的迁移学习方法,微调PVC管道漏水检测模型(1D_CNN)的卷积层数目、激活函数、池化层大小、学习率与优化器等参数与结构,使已有的检测模型(1D_CNN)学习铸铁管漏水数据的特征分布,检测铸铁管是否存在漏水情况。实验使用的铸铁管道漏水数据集中有204 800个样本,已有的PVC管道漏水数据集中有409 600 000个样本,两者均为时序数值型数据。实验结果显示:基于模型的迁移学习方法,使PVC漏水检测模型(1D_CNN)对铸铁管漏水数据检测的准确率由60%提升至92%,表明提出的方法有效。
To address the problems that leak detection models for pipelines have a weak generalization ability and data missing can be caused due to the difficulty in comprehensively collecting leak data for various types of pipelines, this study designed and proposed an approach to transfer a PVC pipe leak detection model based on one-dimensional convolutional neural network (1D_CNN), so as to detect leaks of cast iron pipes. To improve the generalization ability of 1D_CNN, model-based transfer learning was used to fine-tune its parameters and structures such as the number of convolutional layers, activation function, pooling layer sizes, learning rate, and optimizers. This made the existing detection model 1D_CNN can learn the feature distribution of leak data from cast iron pipes and thus detect their leaks. There were 204 800 samples in the cast iron pipe leak dataset for the experiment, and the existing PVC pipe leak dataset contained 409 600 000 samples, both of which were time-series numerical data. The experimental results showed that the model-based transfer learning approach improved the accuracy of the 1D_CNN in detecting leaks of cast iron pipes from 60% to 92%, indicating the effectiveness of the proposed method. This method can shorten the time required to train new detection models, reduce data dependency, and has generalizability.
漏水检测 / 数据缺失 / 检测模型 / 神经网络 / 迁移学习
leak detection / data missing / detection model / neural network / transfer learning
| [1] |
聂维, 江竹, 刘伯相, |
| [2] |
李旭光. 基于时间序列相似性度量的天然气管道泄漏检测方法研究[D]. 重庆: 重庆大学, 2022. |
| [3] |
罗东浩. 基于深度学习的海底管道泄漏检测方法[D]. 哈尔滨: 哈尔滨工程大学, 2019. |
| [4] |
季舒瑶, 袁飞, 程恩, |
| [5] |
|
| [6] |
王晓霞, 艾兴成, 王涛. 基于实例迁移学习的小样本光伏功率短期预测[J]. 太阳能学报, 2024, 45(6): 325-333. |
| [7] |
|
| [8] |
郭俊锋, 王淼生, 王智明. 基于对不平衡数据集进行二次迁移学习的滚动轴承剥落类故障诊断方法[J]. 上海交通大学学报, 2023, 57(11): 1512-1521. |
| [9] |
王克定, 李敬兆, 石晴, |
| [10] |
|
| [11] |
陈文艺, 许洁, 杨辉, |
| [12] |
左保成.漏水信号数据集的增强处理研究[D].呼和浩特:内蒙古师范大学,2023. |
| [13] |
刘艳鹏, 朱立新, 周永章. 卷积神经网络及其在矿床找矿预测中的应用:以安徽省兆吉口铅锌矿床为例[J]. 岩石学报, 2018, 34(11): 3217-3224. |
| [14] |
姬晓飞, 秦琳琳, 王扬扬. 基于RGB和关节点数据融合模型的双人交互行为识别[J]. 计算机应用, 2019, 39(11): 3349-3354. |
| [15] |
|
| [16] |
赵昭. 基于深度神经网络的肺部肿瘤的检测与分割[D]. 南京: 东南大学, 2020. |
| [17] |
曲建岭, 余路, 袁涛, |
中央引导地方科技发展资助项目“基于‘人工智能+’管道漏水检测系统的实现”(2024ZY0144)
内蒙古自治区研究生教育教学改革资助项目“多维度创新型研究生导师团队建设提升培养研究生创新能力的探索”(JG2024001Z)
/
| 〈 |
|
〉 |