采煤机截割状态识别的深度特征融合网络
Network of deep feature fusion for recognition by shearer on cutting state
胥良 1陈鸿垚 1崔兆一1
作者信息
- 1. 黑龙江科技大学 电气与控制工程学院,哈尔滨 150022
- 折叠
摘要
针对综采工作面开采过程中采煤机截割状态识别存在精度低及速度慢等问题,采用卷积神经网络与双向长短时记忆网络相结合的深度学习模型,融合滚筒截割煤岩时生成的信号特征,识别与分类截割状态.通过特征样本移位与叠加随机高斯白噪声扩充样本数量,基于霍普金斯统计量原理建立评价函数与评价指标,构建出增强的特征数据集;将批归一化机制与ReLU激活函数引入到卷积神经网络模块中,加快对增强数据集浅层特征的分组提取收敛速度,拼接浅层特征数据组至双向长短时记忆网络模块中提取更重要的深层状态信息,识别并分类采煤机截割状态,通过与不同模型的对比试验与消融实验,验证深度特征融合网络模型对数据集学习的效果.结果表明,与常规神经网络模型相比,分类识别准确率高2%,训练迭代运算速度快5%,有效地解决了采煤机截割状态识别精度低与速度慢的问题.
Abstract
This paper attempts to address the low accuracy and slow speed in recognition by shearer on the cutting status in mining process of fully mechanized working face.The study involves using a deep learning model combined with convolutional neural network and bidirectional long short-term memory net-work to integrate the signal features generated by cutting coal and rock with various drum,and identify and classify the cutting status;expanding the sample quantity through feature sample shifting and stacking random Gaussian white noise;establishing evaluation function and evaluation index values based on Hop-kins statistical principle,and construct an enhanced feature dataset;introducing batch normalization mechanism and ReLU activation function into the convolutional neural network module to accelerate the convergence speed of grouping and extracting shallow features from the enhanced dataset;concatenating shallow feature data groups into the bidirectional long short-term memory network module to extract more important deep state information;identifying and classifying the cutting state of the coal mining machine;and verifying and analyzing the effectiveness of deep feature fusion network models in dataset learning through comparative experiments and ablation experiments with different models.The results show that compared with conventional neural network models,the accuracy of classification recognition is higher by 2%,and the training iteration speed is faster by 5%,which effectively solves the low accuracy and slow speed in recognizing by the shearer on cutting state.
关键词
采煤机/截割状态/深度学习/数据增强/特征融合Key words
coal mining machine/cutting state/deep learning/data augmentation/feature fusion引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
维普
万方数据