检索结果-南通市图书馆

基于时序分析及CNN-BiLSTM-AM的阶跃型滑坡位移预测

同方期刊数据库博看期刊评论

在线全文

学校读者我要写书评

暂无评论

国外电子测量技术 2024年第1期43卷 126-134页

作者：杨进昆党建武杨景玉岳彪兰州交通大学电子与信息工程学院兰州730070

传统基于递归神经网络的模型对阶跃型滑坡位移预测能力不足,为解决这一问题,提出一种基于时序分析及卷积神经网络-双向长短期记忆-注意力机制(CNN-BiLSTM-AM)的滑坡位移动态预测模型。首先使用变分模态分解方法(VMD)将序列分解为趋势项... 详细信息

传统基于递归神经网络的模型对阶跃型滑坡位移预测能力不足,为解决这一问题,提出一种基于时序分析及卷积神经网络-双向长短期记忆-注意力机制(CNN-BiLSTM-AM)的滑坡位移动态预测模型。首先使用变分模态分解方法(VMD)将序列分解为趋势项、周期项和随机项。采用二次指数平滑法拟合趋势项位移,然后引入最大互信息系数法(MIC)计算各类影响因子与周期项位移相关性,对于周期项和随机项位移采用CNN-BiLSTM-AM混合模型进行多因素和单因素预测,最终累加各分量预测值得到累积位移预测结果。实验结果表明,所提方法在最终累计位移预测结果中拟合系数R~2达0.984和0.987,平均绝对误差(MAE)分别为5.334和3.947,均方根误差(RMSE)分别为6.196和4.941,相比卷积神经网络-长短期记忆(CNN-LSTM)、麻雀搜索算法-核极限学习机(SSA-KELM)和NARX方法,所提方法能够更好的捕捉监测数据的时间相关性,预测精度显著提高,可为阶跃型滑坡预警及防治工作提供参考。

关键词：阶跃型滑坡变分模态分解注意力机制卷积神经网络双向长短时记忆

基于时序序列分解和IBAS LSTM的滑坡数据预测模型

在线全文

学校读者我要写书评

暂无评论

兰州交通大学学报 2024年第2期43卷 58-67页

作者：荆严飞党建武王阳萍岳彪兰州交通大学电子与信息工程学院兰州730070 兰州交通大学轨道交通信息与控制国家级虚拟仿真实验教学中心兰州730070

针对传统静态机器学习模型在周期项位移预测中的缺陷和动态神经网络超参数人工选择困难的问题,在时序序列分解的基础上,提出一种新的滑坡预测耦合模型。首先,用最大相关最小冗余算法对周期项位移筛选合适的环境特征,作为长短期记忆人工... 详细信息

针对传统静态机器学习模型在周期项位移预测中的缺陷和动态神经网络超参数人工选择困难的问题,在时序序列分解的基础上,提出一种新的滑坡预测耦合模型。首先,用最大相关最小冗余算法对周期项位移筛选合适的环境特征,作为长短期记忆人工神经网络的输入。然后,在天牛须搜索算法搜索过程中引入反馈机制,以避免原算法中出现远离最优解的问题;在算法迭代过程中将固定的递减因子改为动态递减因子,以提升前期全局和后期局部的寻优能力;利用改进的天牛须搜索算法对长短期记忆人工神经网络超参数进行寻优,以获得最佳的网络参数组合。最后,重构趋势项和周期项预测结果,得到最终预测位移。以发耳滑坡为例进行分析,结果表明:相较于其他方法,所提模型在平均绝对误差、均方根误差以及拟合度等方面更具优势。

关键词：动态神经网络模型时序序列分解灰色模型长短期记忆人工神经网络天牛须搜索算法

在线全文

学校读者我要写书评

暂无评论

基于生成对抗机制的多目标跟踪方法

兰州交通大学学报 2024年第4期43卷 87-97页

作者：孙逸凡代素敏党建武雍玖兰州交通大学电子与信息工程学院兰州730070 兰州交通大学轨道交通信息与控制国家级虚拟仿真实验教学中心兰州730070 北京中电飞华通信有限公司北京100700

针对多目标跟踪中的跟踪结果易受遮挡而产生的漏检问题,提出一种公路场景下的基于生成对抗机制的多目标跟踪方法。首先,对经过预训练的跟踪网络输出的特征进行处理,在特征空间中添加自适应的二维遮挡掩码,生成现实生活中难以获得的遮挡... 详细信息

针对多目标跟踪中的跟踪结果易受遮挡而产生的漏检问题,提出一种公路场景下的基于生成对抗机制的多目标跟踪方法。首先,对经过预训练的跟踪网络输出的特征进行处理,在特征空间中添加自适应的二维遮挡掩码,生成现实生活中难以获得的遮挡样本;其次,利用生成对抗网络在无监督学习方面的优势,将FairMOT模型作为判别网络,加入结合强化学习机制的生成网络来学习如何筛选困难样本,2个网络进行对抗训练以提升多目标跟踪模型的遮挡不变性,提高跟踪精度;最后,在重识别分支中引入中心损失函数以提高重识别准确度。取数据集BDD100K中部分视频序列进行实验,实验结果表明:改进后算法的跟踪准确率提升了0.8个百分点,跟踪精度降低了0.4个百分点,跟踪过程中身份的切换次数减少了208。

关键词：图像处理多目标跟踪深度学习卷积神经网络生成对抗网络

Inverse Estimation From Speech Sounds To Articulatory Targets

在线全文

学校读者我要写书评

暂无评论

Inverse Estimation From Speech Sounds To Articulatory Target...

第五届全国现代语音学学术会议

作者： jianwu dang Kiyoshi HONDA Information Science School Japan Advanced Institute of Science and Technology ATR Information Science Division

We proposed a method to estimate articulatory targets from speech sounds via a phyiologicaladrticualtory model,which potentially clarifies certain problems of current speech recognition *** the estimation, acoustical parameters are considered as a function of the articulatory targets formodel *** target location is estimated based on a compedson of acouedcal parametersbetween real speech sound and synthetic sound corresponding to the *** proposed estimationmethod was evaluated using vowe-to-vowel *** the target speaker of the model, theaverage estimation error was 0.16 cm for articulatory targets, and 1.8% for the four lower *** suggests that our physiological articulatory model can be a valuable tool for the inverseestimation.

关键词：

COMPARATIVE ANALYSIS OF ARTICULATION MOVEMENTS BETWEEN DISORDER AND NORMAL SPEAKERS

在线全文

cnki会议

学校读者我要写书评

暂无评论

COMPARATIVE ANALYSIS OF ARTICULATION MOVEMENTS BETWEEN DISOR...

第十三届全国人机语音通讯学术会议(NCMMSC2015)

作者： Meng Yuan Hongcui Wang jianwu dang Tianjin Key Lab.of Cognitive Computing and Application Tianjin University Japan Advanced Institute of Science and Technology

Speech disorder speakers refer to people whose pronunciation are incorrect due to the brain central or peripheral nervous damage disease. In the clinical manifestation, it usually performs like frozen and cerebral *** damage influences the control of the speech organs without impacting the comprehension ability. Many previous studies are focusing on the speech characteristics rather than the articulation movements. It is necessary to find the mechanism of disorder speakers’ articulation for helping them to do rehabilitation training. In this paper, the analysis of the acoustic parameter and moving trajectory shows how different of the disorder speakers’ tongue movement from the normal speakers’ tongue movement which would affect the shape of oral and further affect the acoustic *** electromagnetic articulography(EMA) procedures capture the motions of external and internal articulators. We construct TORGO database by labeling and collating and integrate the disorder people and normal people’ EMA data to analyze the acoustic and moving trajectory. On the analysis of moving trajectory, we extract the specific phoneme from continuous speech to research the speaker’s articular *** can be seen from the picture of the movement of sensor attached on the tongue,the disorder speakers’ tongue swings back and forth when they pronounce. And they cannot control the tongue stable for a while. On the acoustic analysis, we use the Mel Frequency Cepstrum Coefficient and linear predictive coding to extract feature parameters from the audio data. Onthis basis, we further extract specific phoneme from continuous speech to research the speaker’s acoustic parameters. We mainly adopt the HMM method to analyze the parameters of acoustic model. The experiment results indicate that HMM method is the effectiveness to investigate the pronunciation mechanism of disorder speakers and normal speakers. Besides the second and third formants of disorder speakers are lower tha

关键词： disorder speakers articular movement speech analysis

Semi-supervised community detection on networks with attributes

在线全文

cnki会议

学校读者我要写书评

暂无评论

Semi-supervised community detection on networks with attribu...

第十三届全国人机语音通讯学术会议(NCMMSC2015)

作者： Jin Di Yingkui Wang jianwu dang Tianjin University

The process of community detection will be benefited when we incorporate all the information available from different sources, which includes not only the information of node/link content and networks topology, but also the partial background knowledge. However, none of current works, to our knowledge, have the ability to make use of all the information available, and it seems to be also a big challenge. Here we give a solution based on stochastic modeling and learning for this problem. Furthermore, we model the related twitter data into a network structure with content information, and then use our proposed semi-supervised community detection method to infer the users’ commercial intents for recommendation in electronic commerce.

关键词： complex networks community detection semi-supervised learning attributes intent comprehension

FREQUENCY-DIFFERENTIATED BRAIN NETWORK DYNAMICS IN REAL WORDS AND PSEUDOWORDS PERCEPTION

在线全文

cnki会议

学校读者我要写书评

暂无评论

中国语音学报 2020年第1期 114-122页

作者： Guo Taiyang ZHAO Bin ZHANG Gaoyan dang jianwu UNOKI Masashi Japan Advanced Institute of Science and Technology Ishikawa PrefectureJapan College of Intelligence and Computing Tianjin UniversityTianjinChina Japan Advanced Institute of Science and Technology(JAIST) Tianjin University TianjinChina School of Information Science at JAIST

Brain mechanisms of lexical-semantic processing have been well researched using electroencephalography(EEG)technique with high temporal ***,the detailed brain dynamics regarding spatial connectivity and the spectral characteristics remain to be *** this reason,this study performed frequency-specific effective connectivity analysis for the EEG recordings during the processing of real and *** addition,we introduced f MRI-based network templates into a representational similarity analysis to compare the functional differences between real and pseudowords in different frequency *** results revealed that real words could rapidly activate the brain network for speech perception and complete its comprehension with efficiency,especially when the first syllable of the real word has clear categorical *** contrast,the pseudowords were delayed in the initiation of speech perception and required a longer time span to retrieve its *** frequency-specific analysis showed that the theta,alpha,and beta rhythms contribute more to semantic processing than the gamma *** results showed that semantic processing is frequency-specific and time-dependent on the word categories.

关键词： Speech perception EEG Brain network Dynamic brain network Frequency specified brain network

SEMI-SUPERVISED COMMUNITY DETECTION ON NETWORKS WITH ATTRIBUTES

在线全文

学校读者我要写书评

暂无评论

SEMI-SUPERVISED COMMUNITY DETECTION ON NETWORKS WITH ATTRIBU...

第十三届全国人机语音通讯学术会议(NCMMSC2015)

作者： Di Jin Yingkui Wang jianwu dang School of Computer Science and Technology Tianjin University Tianjin Key Lab.of Cognitive Computing and Application Tianjin University

Most complex systems in various fields, such as social networks in social science, the Internet in engineering, and signaling pathways in biology, can be formulated as networks where nodes represent entities and links represent some relationship between nodes. Individual entities in a complex system seldom exist in isolation, but rather are often organized in groups to exert functions. For example, an organization is typically consisted of units of different but related functions that interconnect in particular structures to maximize the overall performance of the organization. In biology, a group of proteins in a cell interact to form an RNA polymerase for transcription of genes. Therefore, a critical step toward understanding complex systems is to uncover organizational or community structures in the networks [1]. Communities, also referred to as clusters or modules, are groups of nodes that share common properties or play similar roles. A primary objective of community detection is to identify sets of nodes with common functions by using all the information available in the network. Many methods for community detection have been proposed, and most of them only make use of information of network topology. However, since these methods rely on the accuracy of the relational networks, their ability to discover the true community structures degrades rapidly as the networks are perturbed by noise. Further, this problem will become more complicated when the networks contain complex structures, such as the hierarchical and/or overlapping communities. The above problems may be mitigated when one incorporates some additional knowledge from different sources beyond just network topology. Firstly, a lot of content on nodes and/or edges is often available in many real applications, e.g. Twitter, Flickr, and Facebook in social media. The performance of community detection will be significantly improved if one considers this content information, especially when the network has

关键词： complex networks community detection semi-supervised learning attributes intent comprehension

A GPS/BDS dual-mode positioning algorithm for a train based on CIPSO_EKF

在线全文

cnki会议

学校读者我要写书评

暂无评论

Journal of Measurement Science and Instrumentation 2022年第1期13卷 12-20页

作者： LUO Miao dang jianwu School of Automation and Electrical Engineering Lanzhou Jiaotong University Lanzhou 730070 China Institute of Railway Technology Lanzhou Jiaotong University Lanzhou 730070 China

When using global positioning system/BeiDou navigation satellite(GPS/BDS)dual-mode navigation system to locate a train,Kalman filter that is used to calculate train position has to be adjusted according to the features of the dual-mode *** to multipath effect,positioning accuracy of present Kalman filter algorithm is really *** solve this problem,a chaotic immune-vaccine particle swarm optimization_extended Kalman filter(CIPSO_EKF)algorithm is proposed to improve the output accuracy of the Kalman *** chaotic mapping and immunization,the particle swarm algorithm is first optimized,and then the optimized particle swarm algorithm is used to optimize the observation error covariance *** optimal parameters are provided to the EKF,which can effectively reduce the impact of the observation value oscillation caused by multipath effect on positioning *** the same time,the train positioning results of EKF and CIPSO_EKF algorithms are *** eastward position errors and velocity errors show that CIPSO_EKF algorithm has faster convergence speed and higher real-time performance,which can effectively suppress interference and improve positioning accuracy.

关键词： global positioning system/BeiDou navigation satellite(GPS/BDS)dual-mode positioning chaotic immune-vaccine particle swarm optimization(CIPSO) extended Kalman filter(EKF) positioning accuracy