基于并行禁忌神经网络和DS证据的飞机燃油系统故障诊断

飞机燃油系统是一个由许多相互联系的子系统构成的复杂总体,因而易于发生各类故障,当故障发生时会造成严重影响,为此,设计了一种基于禁忌神经网络和DS证据的飞机燃油系统故障诊断方法;首先,建立了飞机燃油系统的故障诊断模型,然后,建立了3层的BP神经网络故障诊断模型,并采用禁忌优化算法对BP神经网络进行参数优化,得到多个并行诊断的禁忌神经网络,输入样本数据对其训练并利用BP反向传播算法再次调优;最后将测试样本数据输入各禁忌神经网络,并将诊断结果作为证据采用DS证据理论进行融合,得到最终的故障诊断结果;实验结果表明:引入DS证据理论的故障诊断方法能有效克服单一故障诊断方法无法精确诊断故障的不足,诊断精度高,具有较大的优越性。     

粒子群算法优化相空间重构参数的网络流量预测模型

在网络流量预测过程中,相空间重构参数是影响预测性能的重要方面,传统参数分开优化,为了提高网络流量的预测精度,提出一种粒子群算法优化相空间重构参数的网络流量预测模型(PSO-BPNN);首先将BP神经网络作为学习算法,然后采用粒子群算法对相空间重构参数—延迟时间和嵌入维进行联合优化,并重构网络流量序列,最后以小波BP神经网络建立最优络流量预测模型,并采用仿真实验对模型性能进行分析,结果表明,PSO-BPNN提高了网络流量的预测精度。     

基于邻域粗糙集和并行神经网络的故障诊断

针对目前使用神经网络诊断故障时出现的输入向量选择困难、网络结构复杂、对并发故障诊断效果不好等问题,提出了基于邻域粗糙集和并行神经网络的故障诊断方法。先利用邻域粗糙集对初始征兆进行约简,留下有价值的征兆作为神经网络的输入向量,然后针对每种故障类型设计一个神经网络。用多个训练好的神经网络来并行诊断故障,综合每个神经网络的结果给出最终的诊断结论。用转子实验台的实验数据对这种故障诊断方法进行验证,结果显示该方法能优化神经网络结构,且神经网络具有训练速度快、诊断正确率高的特点。     

基于Cholesky分解的增量式RELM及其在时间序列预测中的应用

针对应用于混沌时间序列预测的正则极端学习机(RELM)网络结构设计问题,提出一种基于Cholesky分解的增量式RELM训练算法.该算法通过逐次增加隐层神经元的方式自动确定最佳的RELM网络结构,并以Cholesky分解方式计算其输出权值,有效减小了隐层神经元递增过程的计算代价.混沌时间序列预测实例表明,该算法可有效实现最佳RELM网络结构的自动确定,且计算效率高.利用该算法训练后的RELM预测模型具有预测精度高的优点,适用于混沌时间序列预测. 关键词： 神经网络 极端学习机 混沌时间序列 时间序列预测     

粒子群优化BP神经网络在甲烷检测中的应用

为了准确、快速地检测和预测甲烷气体的浓度,设计了基于红外差分吸收法的甲烷浓度检测系统.为了降低系统部件不稳定带来的影响,检测系统采用双气室结构,气室的输入和输出接口处通过渐变折射率透镜连接到传输光纤,以降低光强的损耗.系统对甲烷检测结果的平均误差为0.007 5.基于粒子群优化的误差反向传播神经网络算法构建了甲烷预测模型,以浓度在0.2%～2.0%范围内的甲烷气体为研究对象.在样本训练过程中,预测模型的精度达到10-4,实际输出值与期望值线性回归的相关系数为0.998 8,最大相对标准偏差为0.248%.实验结果表明,在甲烷浓度预测中,相对于误差反向传播神经网络预测模型,粒子群优化误差反向传播神经网络的预测性能更优.     

应用可见-近红外光谱技术进行白醋品牌和pH值的快速检测

提出了一种基于可见-近红外透射光谱技术快速判别白醋品牌和测定pH值的方法。应用可见-近红外透射光谱获取不同品牌白醋的透射光谱曲线,并对获得的原始光谱数据进行平滑、变量标准化以及一阶导数等预处理,然后利用主成分分析对原始光谱数据进行聚类分析,根据主成分的累计贡献率选取主成分数,并将所选取的主成分作为三层BP神经网络的输入。通过定标集样本对BP神经网络进行训练,得到三层优化神经网络结构：5输入层节点,6隐含层节点和2输出层节点,各层传递函数均采用Sigmoid函数。利用该模型对预测集样本进行预测。实验结果表明在阈值设定为±0.1的情况下该模型对预测集样本品牌鉴别准确率达到了100%,pH预测值与实际测量值偏差小于5%,得到了理想的结果。所以利用可见-近红外光谱技术结合主成分分析和神经网络算法能够快速准确的判定白醋品牌和pH值。     

含有误差校正的小波神经网络交通流量预测

交通流量的准确预测对于高速路管理者进行决策至关重要。建立了小波神经网络(WNN)交通流量预测模型,并通过预测训练误差和测试误差校正预测结果来提高预测精度。首先构建WNN模型对交通流量进行初步预测,然后利用经验模态分解(EMD)和WNN模型对训练误差和测试误差进行预测。分别用训练误差预测值、测试误差预测值和两种误差预测值的加权对流量初步预测结果进行修正得到最终预测值。采用四川省成灌高速路交通流量数据进行了仿真对比实验,仿真结果表明含有误差校正的小波神经网络模型能有效提高交通流量预测精度,并且利用两种误差加权修正模型的预测精度高于利用测试误差的修正模型和利用训练误差的修正模型。     

基于改进教学优化算法的Hermite正交基神经网络混沌时间序列预测

针对传统预测模型对混沌时间序列预测精度低、收敛速度慢及模型结构复杂的问题, 提出了基于改进教学优化算法的Hermite正交基神经网络预测模型. 首先, 将自相关法和Cao方法相结合对混沌时间序列进行相空间重构, 以获得重构延迟时间向量; 其次, 以Hermite正交基函数为激励函数构成Hermite正交基神经网络, 作为预测模型; 最后, 将模型参数优化问题转化为多维空间上的函数优化问题, 利用改进教学优化算法对预测模型进行参数优化, 以建立预测模型并进行预测分析. 分别以Lorenz 系统和Liu系统为模型, 通过四阶Runge-Kutta法产生混沌时间序列作为仿真对象, 并进行单步及多步预测对比实验. 仿真结果表明, 与径向基函数神经网络、回声状态网络、最小二乘支持向量机及基于教学优化算法的Hermite正交基神经网络预测模型相比, 所提预测模型具有更高的预测精度、更快的收敛速度和更简单的模型结构, 验证了该模型的高效性, 便于推广和应用.     

可见/近红外光谱快速测定土壤中的有机碳含量和阳离子交换量

为实现土壤中有机碳(TOC)含量和阳离子交换量(CEC)的快速检测, 对300个土壤样品的可见/近红外光谱数据进行了分析。使用快速独立分量分析(FastICA)算法对光谱数据矩阵进行分解, 得到独立成分和相应的混合系数矩阵, 再利用误差反向传播算法(back-propagation, BP)构造三层神经网络结构。为了克服传统BP神经网络结构难以确定和易于陷入局部极小点的缺点, 采用遗传算法优化BP神经网络结构和初始权值, 得到ICA-GA-BP模型。利用此模型对土壤中TOC含量和CEC进行预测, 根据预测相关系数(R2)和预测标准偏差(RMSEP)来评价预测模型的性能, 表明该模型对TOC含量和CEC测定的相关系数R2均达到0.98以上。说明文章提出的ICA-GA-BP建模方法具有很好的预测效果, 为土壤品质的鉴别提供了一种新方法。     

基于PSO-CNN的土壤氮含量可见/近红外光谱建模

基于机器学习和深度学习的土壤元素含量光谱法建模是土壤化学成分检测的研究热点。为提高基于卷积神经网络的土壤元素含量光谱法建模的准确率,提出了一种粒子群算法优化的卷积神经网络土壤氮元素含量光谱分析模型(PSO-CNN)。对土壤样本进行平滑及标准正态变换处理,减小噪声对建模的影响,设计适用于回归的卷积神经网络结构,采用粒子群算法对设计的卷积神经网络的卷积核参数、学习率、迭代次数等超参数进行优化。可见/近红外光谱的波长范围为225～975nm,对青岛地区的177组土壤样本的氮含量分析建模。建模结果表明,相比偏最小二乘回归、卷积神经网络等建模方法,提出的PSO-CNN建模方法具有更高的预测准确率,测试集的决定系数为0.9707,均方根误差为0.8818,相对分析误差为5.88。     

Guided-Wave Two-Dimensional Acousto-Optic Scanner Using Proton-Exchanged Lithium Niobate Waveguide

Two-dimensional scanning of a 0.6328 mum guided-light beam has been realized using noncolinear acousto-optic (AO) coplanar Bragg diffraction together with colinear AO guided-mode to substrate radiation-mode conversion in a Z-cut Xpropagation LiNbO3 proton-exchanged (PE) waveguide. The two surface acoustic (SAW) waves utilized are at the center frequencies of 500 and 200 MHz, propagating in the Y and X axes, respectively. Two-dimensional scanning of approximately 720 resolvable light beam spots, namely, 18 40 (horizontal vertical) scanning, has been demonstrated using a light beam of 1.0 mm aperture. The total number of resolvable beam spots can be greatly increased from 720 by simply utilizing SAW transducers of larger bandwidth and a light beam of greater aperture. It should also be possible to significantly increase the diffraction efficiency from 3 % by optimizing the parameters of the PE waveguide and the SAWs.     

Hybrid Wavelength and Code Division Multiple Access in Optical Networks

A review of schemes for multiple access in fiber optic networks shows that a hybrid of wavelength and code division multiple access (WCDMA) combines the best features of both. In particular, the hybrid scheme retains the large information carrying capacity of wavelength division multiple access (WDMA) and flexibility of code division multiple access (CDMA). In this paper WDMA, optical CDMA (OCDMA), and WCDMA networks are discussed. In OCDMA networks, concept of incoherent and coherent coding including inverse decoding and matched filter is introduced. The delay performance of both WDMA and WCDMA networks, under the simple suboptimum access protocols based on cyclic search, is computed. It has been shown quantitatively that tuning delay significantly affects the delay performance of both WDMA and WCDMA networks. Futhermore, delay performance of WCDMA networks is always better than the WDMA networks for the same tuning delay, load, and number of users.     

Effect ofγ-radiation on the spectral luminescence characteristics of impure cadmium tungstate crystals

Results of a study of the effect of γ-radiation on the spectral luminescence properties of cadmium tungstate crystals doped with silver, bismuth, and molybdenum cations are presented. Spectral characteristics of the nondnnnoped crystals are briefly described. Absorption and photo and X-ray luminescence spectra of the crystals taken before and after exposure to γ-radiation (5.5·10⁴ Gy) are compared. It is found that the spectral characteristics of the crystals doped with silver, bismuth, and molybdenum cations do not change markedly after the exposure. The relation between the type of impurity-induced defects, individual characteristics of the impurity cations, and the character of the effect of γ-radiation on the spectral luminescence properties of impure crystals is analyzed (preliminarily). Translated from Zhurnal Prikladnoi Spektroskopii, Vol. 64, No. 1, pp. 55–60, January–February, 1997.     

Femtosecond Transform-Limited Pulse Generation by Compensating for the Linear Chirp of SPM Spectra in Dispersion Shifted Fibers

A simple technique of pulse compression, based on the linear chirp compensation of self-phase modulation (SPM) spectra in dispersion shifted fibers, is demonstrated. The optimization procedure is carried out, for a short span of a single-mode fiber, using a parabolic law, which describes the behavior of the squared output pulse width versus the pump peak power in the case of Gaussian pulses. The experimental results give a minimum pulse duration of 233 fs, which is in good agreement with the model. Shorter and coherent pulses, down to 90 fs, have been obtained by inserting an interference filter at the optical output.     

Sorting of bending magnets for the SSRF booster

The Shanghai Synchrotron Radiation Facility(SSRF)booster ring,a full energy injector for the storage ring,is deigned to accelerate the electron beam energy from 150MeV to 3.5GeV that demands high extraction efficiency at the extraction energy with low beam loss rate when electrons are ramping.Closed orbit distortion(COD)caused by bending magnet field uniformity errors which affects the machine performance harmfully could be effectively reduced by bending magnet location sorting.Considering the affections of random errors in measurement,both ideal sorting and realistic sorting are studied based on measured bending magnet field uniformity errors and one reasonable combination of bending magnets which can reduce the horizontal COD by a factor of 5is given as the final installation sequence of the booster bending magnets in this paper.     

Measurement of nonlinear coefficient of optical fiber based on small chirped soliton transmission

We measure the waveform and phase curves of short optical pulses before and after transmission over different lengths of fibers by use of the pulse analyzer with the frequency-resolved optical gating (FROG),and numerically simulate pulse evolution under the experimental conditions.The nonlinear coefficient of the fiber is given by comparing the experimental results with the numerical ones.Difference between the experiment and numerical simulation is analyzed.     

Resonance scattering of canonical elastic shells in absorbing fluid medium

Resonance scattering of elastic spherical shell and cylindrical shell while the surrounding fluid medium has absorption is studied. The normal mode solution derived using exact elastic theory and the separation of variables is still applicable. However, the scattering form function has to be modified for the absorbing medium, otherwise the unreasonable result would be obtained. The backscattering form function in the absorbing medium is redefined, and the form function of elastic spherical and cylindrical shell with vacuum or solid matter filled is calculated in various absorption conditions. The results show that the absorption of surrounding fluid leads to notable attenuation of the coincidence resonances in the mid-frequency, but it has a little influence on the low-frequency resonance scattering induced by the filler inside the shell.     

Single-beam self-referenced phase-sensitive surface plasmon resonance sensor with high detection resolution

A versatile and low-cost single-beam self-referenced phase-sensitive surface plasmon resonance(SPR)sensing system with ultra-high resolution performance is presented.The system exhibits a root-mean-square phase fluctuation of ±0.0028.over a period of 45 min.i.e.a resolution of±5.2×10-9 refractive index units.The enhanced performance has been achieved through the incorporation of three design elements:a true single-beam configuration enabling complete self-referencing so that only the phase change associated with SPR gets detected,a differential measurement scheme to eliminate spurious signals not related to the sensor response,and the elimination of retardation drifts by incorporating temperature stabilization in the liquid crystal phase modulato .Our design should bring the detection sensitivity of non-labeling SPR biosensing closer to that achievable by conventional fluorescence-based techniques.     

Influences of SiO2 protective layers and annealing on the laser-induced damage threshold of Ta2O5 films

Ta2O5 films are prepared on BK7 substrates with conventional electron beam evaporation deposition.The effects of SiO2 protective layers and annealing on the laser-induced damage threshold (LIDT) of the films are investigated.The results show that SiO2 protective layers exert little influence on the electric field intensity(EFI)distribution,microstructure and microdefect density but increase the absorption slightly.Annealing iS effective on decreasing the microdefect density and the absorption of the films.Both SiO2 protective layers and annealing are beneficial to the damage resistance of the films and the latter is more effective to improve the LIDT.Moreover,the maximal LIDT of Ta2O5 films is achieved by the combination of SiO2 protective layers and annealing.