摘要
煤矿井下复杂的地质环境和强电磁干扰,导致钻孔雷达信号信噪比低、有效信息提取困难,严重制约地质构造探测精度。传统滤波方法因模型假设局限性和自适应能力不足,难以有效应对井下非平稳、非线性噪声干扰。本文提出一种基于卷积神经网络计算技术的智能化滤波方法,融合自适应变分模态分解方法,实现噪声与有效信号的高效分离。通过自适应变分模态分解对原始信号进行本征模态分解,抑制高频噪声干扰;构建多尺度注意力残差网络,提取信号时频域深层特征,增强有效反射波的边缘信息保留能力。实验结果表明,与传统小波滤波和卡尔曼滤波方法相比,本文方法在信噪比和均方根误差指标上均有效有提升,计算效率满足井下实时处理需求。
The complex geological environment and strong electromagnetic interference underground in coal mines result in low signal-to-noise ratio of drilling radar signals and difficulty in extracting effective information,seriously restricting the accuracy of geological structure detection.Traditional filtering methods are difficult to effectively deal with non-stationary and nonlinear noise interference underground due to limitations in model assumptions and insufficient adaptability.This paper propose an intelligent filtering method based on convolutional neural network computing technology,integrating adaptive variational mode decomposition method to achieve efficient separation of noise and effective signals.Perform intrinsic mode decomposition on the original signal through adaptive variational mode decomposition to suppress high-frequency noise interference.Constructing a multi-scale attention residual network to extract deep time-frequency features of signals and enhance the ability to preserve edge information of effective reflected waves.The experimental results show that compared with traditional wavelet filtering and Kalman filtering methods,the proposed method effectively improves the signal-to-noise ratio and root mean square error indicators,and the computational efficiency meets the real-time processing requirements underground.
作者
张军
张鹏
赵朋朋
王霄菲
邓立博
王智聪
Zhang Jun;Zhang Peng;Zhao Pengpeng;Wang Xiaofei;Deng Libo;Wang Zhicong(CCTEG Xi'an Research Institute(Group)Co.,Ltd.,Xi'an 710077,China)
出处
《煤炭技术》
2026年第1期196-200,共5页
Coal Technology
关键词
卷积神经网络
钻孔雷达
智能化滤波
深度学习
杂波抑制
convolutional neural network
drilling radar
intelligent filtering
deep learning
clutter suppression
