Marine oil spills have caused major threats to marine environment over the past few years.The early detection of the oil spill is of great significance for the prevention and control of marine disasters.At present,rem...Marine oil spills have caused major threats to marine environment over the past few years.The early detection of the oil spill is of great significance for the prevention and control of marine disasters.At present,remote sensing is one of the major approaches for monitoring the oil spill.Full polarization synthetic aperture radarc SAR data are employed to extract polarization decomposition parameters including entropy（H） and reflection entropy（A）.The characteristic spectrum of the entropy and reflection entropy combination has analyzed and the polarization characteristic spectrum of the oil spill has developed to support remote sensing of the oil spill.The findings show that the information extracted from（1-A）×（1-H） and（1-H）×A parameters is relatively evident effects.The results of extraction of the oil spill information based on H×A parameter are relatively not good.The combination of the two has something to do with H and A values.In general,when H〉0.7,A value is relatively small.Here,the extraction of the oil spill information using（1-A）×（1-H） and（1-H）×A parameters obtains evident effects.Whichever combined parameter is adopted,oil well data would cause certain false alarm to the extraction of the oil spill information.In particular the false alarm of the extracted oil spill information based on（1-A）×（1-H） is relatively high,while the false alarm based on（1-A）×H and（1-H）×A parameters is relatively small,but an image noise is relatively big.The oil spill detection employing polarization characteristic spectrum support vector machine can effectively identify the oil spill information with more accuracy than that of the detection method based on single polarization feature.展开更多
In cognitive radio networks, Secondary Users (SUs) have opportunities to access the spectrum channel when primary user would not use it, which will enhance the resource utilization. In order to avoid interference to p...In cognitive radio networks, Secondary Users (SUs) have opportunities to access the spectrum channel when primary user would not use it, which will enhance the resource utilization. In order to avoid interference to primary users, it is very important and essential for SUs to sense the idle spectrum channels, but also it is very hard to detect all the channels in a short time due to the hardware restriction. This paper proposes a novel spectrum prediction scheme based on Support Vector Machines (SVM), to save the time and energy consumed by spectrum sensing via predicting the channels' state before detecting. Besides, spectrum utilization is further improved by using the cooperative mechanism, in which SUs could share spectrum channels' history state information and prediction results with neighbor nodes. The simulation results show that the algorithm has high prediction accuracy under the condition of small training sample case, and can obviously reduce the detecting energy, which also leads to the improvement of spectrum utilization.展开更多
针对齿轮箱振动信号复杂多变,导致现有的齿轮箱故障诊断方法诊断精度不高、较弱故障特征容易被噪声淹没等问题,提出了一种基于向量加权平均优化算法(weighted mean of vectors,INFO)、变分模态分解(variational mode decomposition,VMD...针对齿轮箱振动信号复杂多变,导致现有的齿轮箱故障诊断方法诊断精度不高、较弱故障特征容易被噪声淹没等问题,提出了一种基于向量加权平均优化算法(weighted mean of vectors,INFO)、变分模态分解(variational mode decomposition,VMD)和卷积神经网络(convolutional neural network,CNN)的齿轮故障诊断方法。该方法首先采用熵权法将不同位置的振动传感器信号信息进行融合,利用INFO对VMD算法中参数进行优化,并设计一个复合评价指标作为参数优化的评价标准,使用奇异峭度差分谱的方法对敏感分量进行重构;其次,从重构的信号中提取时域、频域特征并输入到CNN模型中进行分类;最后通过Shap(Shapley additive explanations)值法对模型输入特征的重要性进行排序,分析不同特征组合对模型分类和特定故障识别的影响。在东南大学行星齿轮数据集上进行验证,结果表明,利用所提特征组合进行故障诊断,CNN模型故障诊断准确率为98.24%,高于其他特征组合,为行星齿轮箱的故障诊断提供了一组有效的特征指标。展开更多
针对传统频谱感知算法在复杂信道环境下鲁棒性欠佳的问题,以及深度学习感知算法面临的模型训练复杂度高等局限,提出了一种融合多种群人工鱼群算法与模糊孪生支持向量机(fuzzy twin support vector machine,FTSVM)的频谱感知方法.首先,...针对传统频谱感知算法在复杂信道环境下鲁棒性欠佳的问题,以及深度学习感知算法面临的模型训练复杂度高等局限,提出了一种融合多种群人工鱼群算法与模糊孪生支持向量机(fuzzy twin support vector machine,FTSVM)的频谱感知方法.首先,通过计算接收信号协方差矩阵的迹及其对角线外元素的均值,构建一个二维特征向量,由FTSVM进行训练识别;然后,使用样本的模糊隶属度调整了FTSVM超平面,从而使训练得到的模型更倾向于识别出初级用户存在的信号;最后,设计了多种群机制的改进人工鱼群算法,对频谱感知模型参数进行优化.仿真实验结果表明,在面临小样本数据集和低信噪比环境时,所提方法相较于其它的特征提取和SVM方法,在模型感知性能上实现了有效提升,−20 dB信噪比下检测概率达0.7以上.同时,优化算法的多种群机制缩短了模型的训练时间,相较于改进人工鱼群算法,训练时间缩短了约81%.展开更多
为更全面提取转子故障特征,将全矢谱和复局部均值分解(Complex local mean decomposition,CLMD)相结合,提出二元的全矢包络技术——CLMD全矢包络技术。采用正交采样技术获取转子同一截面上互相垂直方向上的振动信号,并将其组成一个复数...为更全面提取转子故障特征,将全矢谱和复局部均值分解(Complex local mean decomposition,CLMD)相结合,提出二元的全矢包络技术——CLMD全矢包络技术。采用正交采样技术获取转子同一截面上互相垂直方向上的振动信号,并将其组成一个复数信号;运用CLMD将复数信号按能量从高到低的顺序依次分离出系列复乘积函数(Complex product function,CPF),并解调出CPF的复包络;由于故障特征主要在能量较高的CPF分量中,通过全矢谱技术融合前几阶CPF分量的包络信号,得到相应的全矢包络谱。仿真的调幅-调频信号分析结果表面,较之Hilbert解调,CLMD全矢包络技术可提取隐含的调频信息,而且不存在虚假的低频谱线。转子试验台模拟的基座松动信号、碰摩信号分析结果表明,较之单源信息的包络谱,CLMD全矢包络技术提取的谱线特征更清晰、全面,而且根据全矢包络谱可有效区分基座松动引起的碰摩和单一碰摩故障。展开更多
基金The National Natural Science Foundation of China under contract No.41376183the Oceanography Public Welfare Scientific Research Project "Marine oil spill risk assessment and key technologies of emergency response integration and demonstration" under contract No.201205012
文摘Marine oil spills have caused major threats to marine environment over the past few years.The early detection of the oil spill is of great significance for the prevention and control of marine disasters.At present,remote sensing is one of the major approaches for monitoring the oil spill.Full polarization synthetic aperture radarc SAR data are employed to extract polarization decomposition parameters including entropy（H） and reflection entropy（A）.The characteristic spectrum of the entropy and reflection entropy combination has analyzed and the polarization characteristic spectrum of the oil spill has developed to support remote sensing of the oil spill.The findings show that the information extracted from（1-A）×（1-H） and（1-H）×A parameters is relatively evident effects.The results of extraction of the oil spill information based on H×A parameter are relatively not good.The combination of the two has something to do with H and A values.In general,when H〉0.7,A value is relatively small.Here,the extraction of the oil spill information using（1-A）×（1-H） and（1-H）×A parameters obtains evident effects.Whichever combined parameter is adopted,oil well data would cause certain false alarm to the extraction of the oil spill information.In particular the false alarm of the extracted oil spill information based on（1-A）×（1-H） is relatively high,while the false alarm based on（1-A）×H and（1-H）×A parameters is relatively small,but an image noise is relatively big.The oil spill detection employing polarization characteristic spectrum support vector machine can effectively identify the oil spill information with more accuracy than that of the detection method based on single polarization feature.
基金Sponsored by the Youth Foundation of Beijing Univesity of Postsand Telecommunications(Grant No.2011RC0110)Director Foundation of Key Lab of Universal Wirelsess Communication of Ministry of Education(Grant No.ZRJJ-2010-3)Ministry of Industry and Information Technology of China(Grant No.2011ZX03001-007-03)
文摘In cognitive radio networks, Secondary Users (SUs) have opportunities to access the spectrum channel when primary user would not use it, which will enhance the resource utilization. In order to avoid interference to primary users, it is very important and essential for SUs to sense the idle spectrum channels, but also it is very hard to detect all the channels in a short time due to the hardware restriction. This paper proposes a novel spectrum prediction scheme based on Support Vector Machines (SVM), to save the time and energy consumed by spectrum sensing via predicting the channels' state before detecting. Besides, spectrum utilization is further improved by using the cooperative mechanism, in which SUs could share spectrum channels' history state information and prediction results with neighbor nodes. The simulation results show that the algorithm has high prediction accuracy under the condition of small training sample case, and can obviously reduce the detecting energy, which also leads to the improvement of spectrum utilization.
文摘针对齿轮箱振动信号复杂多变,导致现有的齿轮箱故障诊断方法诊断精度不高、较弱故障特征容易被噪声淹没等问题,提出了一种基于向量加权平均优化算法(weighted mean of vectors,INFO)、变分模态分解(variational mode decomposition,VMD)和卷积神经网络(convolutional neural network,CNN)的齿轮故障诊断方法。该方法首先采用熵权法将不同位置的振动传感器信号信息进行融合,利用INFO对VMD算法中参数进行优化,并设计一个复合评价指标作为参数优化的评价标准,使用奇异峭度差分谱的方法对敏感分量进行重构;其次,从重构的信号中提取时域、频域特征并输入到CNN模型中进行分类;最后通过Shap(Shapley additive explanations)值法对模型输入特征的重要性进行排序,分析不同特征组合对模型分类和特定故障识别的影响。在东南大学行星齿轮数据集上进行验证,结果表明,利用所提特征组合进行故障诊断,CNN模型故障诊断准确率为98.24%,高于其他特征组合,为行星齿轮箱的故障诊断提供了一组有效的特征指标。
文摘为更全面提取转子故障特征,将全矢谱和复局部均值分解(Complex local mean decomposition,CLMD)相结合,提出二元的全矢包络技术——CLMD全矢包络技术。采用正交采样技术获取转子同一截面上互相垂直方向上的振动信号,并将其组成一个复数信号;运用CLMD将复数信号按能量从高到低的顺序依次分离出系列复乘积函数(Complex product function,CPF),并解调出CPF的复包络;由于故障特征主要在能量较高的CPF分量中,通过全矢谱技术融合前几阶CPF分量的包络信号,得到相应的全矢包络谱。仿真的调幅-调频信号分析结果表面,较之Hilbert解调,CLMD全矢包络技术可提取隐含的调频信息,而且不存在虚假的低频谱线。转子试验台模拟的基座松动信号、碰摩信号分析结果表明,较之单源信息的包络谱,CLMD全矢包络技术提取的谱线特征更清晰、全面,而且根据全矢包络谱可有效区分基座松动引起的碰摩和单一碰摩故障。
