基于频谱动态特征和ELM的挖掘设备识别方法研究
发布时间：2018-11-21 19:43
【摘要】：近几年,随着我国城市化建设飞速发展对地下电缆的安全性需求越来越迫切。由于在道路改造和房屋建设等施工过程中,工作人员的疏忽大意导致电缆被挖断的事故频频发生,给国家经济和人民安全带来严重危害。因此,保障地下电缆供电系统不受挖掘设备破坏成为我国电力及城建部门亟待解决的问题。本文在语音识别的基础上,对常用几种挖掘设备(挖掘机、液压冲击锤、电锤、切割机)的声音信号展开深入分析研究,构建了一套基于频谱动态特征的声音信号提取方法和极限学习机(ELM)作为分类器的挖掘设备识别算法。该算法能够有效地检测到威胁电缆安全的挖掘设备在作业时的声音信号,从而进行预警判断,达到对事发地进行定位的目的。本文主要研究工作如下:1.采用八通道的麦克风十字阵列在夜晚较理想的环境下对四种挖掘设备在不同距离作业下采集声音信号,用于建立声音特征库。通过声阵列对不同环境、不同距离下挖掘设备在白天正常作业的声音信号进行采集与识别,进一步验证该算法的有效性。2.采用基于Mel频率倒谱系数(MFCC)特征提取方法、基于一阶差分Mel频率倒谱系数((35)MFCC)特征提取方法、基于二阶差分Mel频率倒谱系数((35)(35)MFCC)特征提取方法和基于频谱动态特征的声音信号提取方法。通过对挖掘设备声音信号的特征提取,进行不同的对比实验。3.在模式识别方面,基于识别率、训练模型和识别时间长短作为本文算法的评价标准。选取BP前馈神经网络、KNN和ELM三种模式识别方法,用于对挖掘设备信号类型的识别对比。4.在实验中,设计了基于MFCC、(35)MFCC、(35)(35)MFCC和频谱动态特征的系数提取以及BP前馈神经网络、KNN和ELM三种分类识别算法的对比实验。进一步讨论了隐含结点个数以及KNN识别算法中K值对识别结果的影响。通过大量实验进行分析,基于频谱动态特征的声音特征提取方法和ELM的识别算法对挖掘设备作业的异常事件识别及预警是稳定的。5.为增强算法的鲁棒性,在地铁施工现场,重新采集挖掘设备声音数据验证每种设备的工作状态。结果表明,该算法能够较准确地对挖掘设备进行识别从而达到预报警的目的。最后将该识别算法通过MATLAB软件建立一个GUI界面。
[Abstract]:In recent years, with the rapid development of urbanization in China, the security needs of underground cables are becoming more and more urgent. In the process of road reconstruction and building construction, the carelessness of the workers leads to frequent accidents of cable breaking, which brings serious harm to the national economy and the safety of the people. Therefore, the protection of underground cable power supply system from excavation equipment damage has become an urgent problem for power and urban construction departments in China. On the basis of speech recognition, the sound signals of several kinds of excavating equipment (excavator, hydraulic hammer, electric hammer, cutting machine) are deeply analyzed and studied in this paper. A set of acoustic signal extraction methods based on spectrum dynamic features and a mining equipment recognition algorithm based on extreme learning machine (ELM) as classifier are constructed. The algorithm can effectively detect the sound signal of the mining equipment which threatens the safety of the cable in the operation, so as to carry out early warning judgment and achieve the purpose of locating the site of the accident. The main work of this paper is as follows: 1. An eight-channel microphone cross array is used to collect sound signals from four kinds of excavating devices at different distances in an ideal environment at night, which can be used to set up a sound signature database. The acoustic array is used to collect and recognize the sound signals of the mining equipment in different environments and at different distances, which further verifies the effectiveness of the algorithm. 2. The (MFCC) feature extraction method based on Mel frequency cepstrum coefficient and the first order differential Mel frequency cepstrum coefficient (35) MFCC) feature extraction method) are used. Second order difference Mel frequency cepstrum coefficients (35) (35) MFCC) feature extraction method and sound signal extraction method based on spectrum dynamic features are presented. Through the feature extraction of the acoustic signal of mining equipment, different contrast experiments are carried out. 3. In the aspect of pattern recognition, recognition rate, training model and recognition time are the evaluation criteria of this algorithm. Three pattern recognition methods, BP feedforward neural network, KNN and ELM, are selected to identify and compare the signal types of mining equipment. 4. In the experiment, the coefficients extraction based on MFCC, (35) (35) MFCC and spectrum dynamic features, as well as three classification and recognition algorithms based on BP feedforward neural network, KNN and ELM are designed. Furthermore, the effect of the number of hidden nodes and the K value in KNN recognition algorithm on the recognition results is discussed. Based on a large number of experiments, the acoustic feature extraction method based on dynamic spectrum features and the recognition algorithm of ELM are stable for the detection and warning of abnormal events in mining equipment operations. In order to enhance the robustness of the algorithm, the acoustic data of each kind of equipment are collected again at the subway construction site to verify the working state of each equipment. The results show that the algorithm can accurately identify the mining equipment and achieve the purpose of pre-warning. Finally, a GUI interface is built by using MATLAB software.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.34;TM75
本文编号：2348087
[Abstract]:In recent years, with the rapid development of urbanization in China, the security needs of underground cables are becoming more and more urgent. In the process of road reconstruction and building construction, the carelessness of the workers leads to frequent accidents of cable breaking, which brings serious harm to the national economy and the safety of the people. Therefore, the protection of underground cable power supply system from excavation equipment damage has become an urgent problem for power and urban construction departments in China. On the basis of speech recognition, the sound signals of several kinds of excavating equipment (excavator, hydraulic hammer, electric hammer, cutting machine) are deeply analyzed and studied in this paper. A set of acoustic signal extraction methods based on spectrum dynamic features and a mining equipment recognition algorithm based on extreme learning machine (ELM) as classifier are constructed. The algorithm can effectively detect the sound signal of the mining equipment which threatens the safety of the cable in the operation, so as to carry out early warning judgment and achieve the purpose of locating the site of the accident. The main work of this paper is as follows: 1. An eight-channel microphone cross array is used to collect sound signals from four kinds of excavating devices at different distances in an ideal environment at night, which can be used to set up a sound signature database. The acoustic array is used to collect and recognize the sound signals of the mining equipment in different environments and at different distances, which further verifies the effectiveness of the algorithm. 2. The (MFCC) feature extraction method based on Mel frequency cepstrum coefficient and the first order differential Mel frequency cepstrum coefficient (35) MFCC) feature extraction method) are used. Second order difference Mel frequency cepstrum coefficients (35) (35) MFCC) feature extraction method and sound signal extraction method based on spectrum dynamic features are presented. Through the feature extraction of the acoustic signal of mining equipment, different contrast experiments are carried out. 3. In the aspect of pattern recognition, recognition rate, training model and recognition time are the evaluation criteria of this algorithm. Three pattern recognition methods, BP feedforward neural network, KNN and ELM, are selected to identify and compare the signal types of mining equipment. 4. In the experiment, the coefficients extraction based on MFCC, (35) (35) MFCC and spectrum dynamic features, as well as three classification and recognition algorithms based on BP feedforward neural network, KNN and ELM are designed. Furthermore, the effect of the number of hidden nodes and the K value in KNN recognition algorithm on the recognition results is discussed. Based on a large number of experiments, the acoustic feature extraction method based on dynamic spectrum features and the recognition algorithm of ELM are stable for the detection and warning of abnormal events in mining equipment operations. In order to enhance the robustness of the algorithm, the acoustic data of each kind of equipment are collected again at the subway construction site to verify the working state of each equipment. The results show that the algorithm can accurately identify the mining equipment and achieve the purpose of pre-warning. Finally, a GUI interface is built by using MATLAB software.
【学位授予单位】：杭州电子科技大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.34;TM75
【相似文献】
相关期刊论文 前4条
1 李富安;;挖掘设备潜力,提高供电质量[J];排灌机械;1986年05期
2 ;挖掘设备潜力 扩大接卸能力[J];哈铁科技通讯;1983年06期
3 马守明;连续墙抓斗挖掘设备及工艺[J];探矿工程(岩土钻掘工程);1995年03期
4 ;[J];;年期
相关会议论文 前1条
1 纪建平;焦惠丽;侯小春;;依靠技术改造 挖掘设备潜能[A];“常纺院”杯全国前织生产技术管理经验交流会论文集[C];2008年
相关重要报纸文章 前3条
1 本报记者 王彦彬;深度挖掘设备节能[N];通信产业报;2013年
2 记者 苏建凤;首钢长钢炼铁厂 降本增效有绝招[N];首都建设报;2010年
3 晓兰、永生、开芳;津西钢铁利润水平位居全国钢铁行业前十[N];唐山劳动日报;2009年
相关硕士学位论文 前1条
1 赵拓;基于频谱动态特征和ELM的挖掘设备识别方法研究[D];杭州电子科技大学;2016年
，本文编号：2348087
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2348087.html
