摘要
为避免地下电缆遭受破坏,提高振动监测系统对外力破坏的预警能力,提出一种基于时频谱图和自适应动态权重粒子群算法-卷积神经网络(PSO-CNN)的外破振动信号识别方法。首先,将振动传感系统获取的3000组外破振动信号转化生成为时频谱图数据集,在图像预处理阶段,采用直方图均衡化和二维主成分分析(2D-PCA)算法来增强灰度图像特征并实现图像数据的降维;然后,将图像数据集的70%作为CNN模型的训练集,并在网络训练过程中引入自适应动态惯性权重PSO对CNN模型的卷积层、池化层相关参数进行迭代寻优,从而获得优化PSO-CNN分类模型;最后,利用测试集图像数据对优化PSO-CNN模型的识别性能进行验证,并与其他分类模型进行了对比。结果表明,所提方法对6种常见外破振动信号的识别准确率达到98.33%,平均每张图像的识别时间仅为0.24 s,与其他分类算法相比具有更高的分类精度和更快速的识别速度,为快速准确地识别外力破坏事件类型提供了一种可行方案。
In order to avoid the damage of underground cables and improve the early warning ability of the vibration monitoring system against external force damage,an identification method of external damage vibration signal based on time-frequency spectrum and adaptive dynamic inertia weight PSO-CNN is proposed.Firstly,3000 groups of external vibration signals obtained by the vibration sensing system are converted into time-frequency spectrum data sets.In the image preprocessing stage,histogram equalization and 2D-PCA algorithm are used to enhance the characteristics of gray image and reduce the dimensions of image data;Then,70%of the image dataset is taken as the training set of the CNN model,and the adaptive dynamic inertia weight PSO algorithm is introduced in the network training process to iteratively optimize the relevant parameters of the convolution layer and pooling layer of the CNN model,so as to obtain the optimized PSO-CNN classification model;Finally,the recognition performance of the optimized PSO-CNN model is verified by using test set image data,and which is compared with other classification models.The results show that the recognition accuracy of the proposed method for six common external damage vibration signals reaches 98.33%,and the average recognition time of each image is only 0.24 s.Compared with other classification algorithms,the proposed method has higher classification accuracy and faster recognition speed,which provides a feasible scheme for quickly and accurately identifying the types of external damage events.
作者
崔岩
方春华
文中
方萌
游海鑫
郭俊康
Cui Yan;Fang Chunhua;Wen Zhong;Fang Meng;You Haixin;Guo Junkang(College of Electricity and New Energy,China Three Gorges University,Yichang 443000,China)
出处
《国外电子测量技术》
北大核心
2023年第1期144-152,共9页
Foreign Electronic Measurement Technology
基金
国家自然科学基金(51477090)项目资助
关键词
时频谱图
2D-PCA降维
惯性权重
卷积神经网络
粒子群优化算法
time-frequency spectrum
2D-PCA dimension reduction
inertia weight
convolutional neural network
particle swarm optimization algorithm
