基于卡尔曼滤波的陀螺仪降噪处理

针对MEMS陀螺仪输出信号随机漂移误差造成测量精度低的问题,提出了一种基于BP神经网络的卡尔曼滤波降噪模型.基于BP神经网络的基本原理,首先利用BP神经网络对系统进行学习,获得系统状态方程,然后建立了基于BP神经网络的滤波模型,最后应用于卡尔曼滤波对MEMS陀螺仪信号进行降噪.半实物模拟仿真实验表明:基于BP神经网络的卡尔曼滤波后的数据的速率随机游走等系数比原始数据下降6.89倍,验证了本方法的降噪性能优于基本卡尔曼模型,在MEMS陀螺仪的数据处理方面具有一定的应用价值.     

卷积神经网络特征重要性分析及增强特征选择模型

卷积神经网络等深度神经网络凭借着其强大的表达能力、突出的分类性能,已在不同领域内得到了广泛应用.当面对高维特征时,深度神经网络通常被认为具有较好的鲁棒性,能够隐含地对特征进行选择,但由于网络参数巨大,如果数据量达不到足够的规模,则会导致学习不充分,因而可能无法达到最优的特征选择.而神经网络的黑箱特性使得无法观测神经网络选择了哪些特征,也无法评估其特征选择的能力.为此,以卷积神经网络为例,首先研究如何显式地表达神经网络中的特征重要性,提出了基于感受野的特征贡献度分析方法;其次,将神经网络特征选择与传统特征评价方法进行对比分析发现,在非海量样本的情况下,传统特征评价方法对高重要性特征和噪声特征的识别能力反而能够超过神经网络.因此,进一步地提出了卷积神经网络增强特征选择模型,将传统特征评价方法对特征重要性的理解结合到神经网络的学习过程中,以辅助深度神经网络进行特征选择.在基于文本的社交媒体用户属性建模任务下进行了对比实验,结果验证了该模型的有效性.     

神经网络修正动态GPS卡尔曼滤波算法研究

GPS导航定位系统噪声具有非先验性,而卡尔曼滤波进行最优估计需建立准确的系统模型和观测模型,这导致标准卡尔曼滤波的精度不高。为提高滤波精度,提出了神经网络修正动态GPS卡尔曼滤波算法,采用两个BP神经网络分别在时间更新预测部分及测量更新部分对标准卡尔曼滤波器进行修正,这样既考虑了现实环境的动态变化对系统模型造成的随机干扰影响,又融合了神经网络的自学习性和自适应性,使其对动态环境的扰动具有了自适应能力。仿真研究表明：该算法优于标准卡尔曼滤波器。     

THE UNIT RBFNETWORK: EXPERIMENTS AND PRELIMINARY RESULTS

URBF, the unit radial basis function network is an RBF neural network with all second layer weights set to - 1. The URBF models functions or physical phenomena by sampling their behaviors at various probe points, and correcting the model, more and more delicately (i.e., using Gaussian functions with ever narrower spread), when discrepancies are discovered. The probe point - input space positions to test and adjust the network - are linear pixel shuffling points, used for their highly uniform sampling property. We demonstrate the network's performance on several examples. It shows its power via good extrapolation behavior: for smooth-boundary discriminations, very few new hidden units need to be added for a large number of probe points.     

少样本文本分类的多任务原型网络

少样本文本分类中,原型网络对语义利用不足、可迁移特征挖掘不够,导致模型泛化能力不强,在新任务空间中分类性能不佳。从模型结构、编码网络、度量网络等角度提高模型泛化性,提出多任务原型网络(multiple-task prototypical network, MTPN)。结构上,基于原型网络度量任务增加辅助分类任务约束训练目标,提高了模型的语义特征抽取能力,利用多任务联合训练,获得与辅助任务更相关的语义表示。针对编码网络,提出LF-Transformer编码器,使用层级注意力融合底层通用编码信息,提升特征的可迁移性。度量网络使用基于BiGRU的类原型生成器,使类原型更具代表性,距离度量更加准确。实验表明,MTPN在少样本文本情感分类任务中取得了91.62%的准确率,比现有最佳模型提升了3.5%以上;在新领域的情感评论中,基于五条参考样本,模型对查询样本可获得超过90%的分类准确率。     

一种基于非线性滤波的倒立摆控制方法研究

对于倒立摆这样的强非线性系统,采用传统的BP算法存在着收敛速度慢、易陷入局部极小值的缺陷,而采用卡尔曼滤波方法则会带来很大的模型误差。为了解决上述问题,提出了基于粒子滤波优化神经网络的方法。首先建立了倒立摆神经网络控制器的物理模型并将模型粒子化,而后用粒子滤波算法对粒子进行优化估计,将估计结果作为网络的权值应用到倒立摆控制中,采用离线训练方式,仿真比较了卡尔曼滤波和粒子滤波两种方法控制效果,结果表明,新算法较卡尔曼滤波方法在控制性能上有明显提高。     

Artificial neural networks applied to classification of mutagenic activity of nitro-substituted polycyclic aromatic hydrocarbons

Three-layer artificial neural networks (ANN) with back-propagation of error have been applied to classification of nitro-substituted polycyclic aromatic hydrocarbons (NPAH) based on the regularity that the structure difference of NPAH compounds leads to different mutagenic activity towards Salmonella typhimurium. The network's architecture and parameters were optimized to give maximum correct classification rate of 93.8% for two different classes of NPAHs: weakly active and strongly active ones. The results compared favourably with those obtained by nonlinear mapping pattern recognition method. Considering that the most important factor for NPAH's mutagenicity might be the electron effect of the substituted nitro-groups, an electrotopological state index of nitrogen atom was introduced additionally as one of network's inputs, and the correct classification rate was consequently raised to 97.5%. The network's prediction ability for untrained samples was also satisfactory.     

Estimating modes of a complex dynamical network from impulse response data: Structural and graph‐theoretic characterizations

We examine the role played by a linear dynamical network's topology in inference of its eigenvalues from noisy impulse‐response data. Specifically, for a canonical linear‐time‐invariant network dynamics, we relate the Cramer–Rao bounds on eigenvalue estimator performance (from impulse‐response data) to structural properties of the transfer function and in turn, to the network's topological structure. We begin by reviewing and enhancing algebraic characterizations of such eigenvalue estimates, which are based on pole‐residue and pole‐zero representations of the network's dynamics. We use these results to characterize mode estimation in networks with slow‐coherency structures, finding that stimulus and observation in each strongly connected network subgraph is needed for high‐fidelity estimation. We also obtain spectral and graphical characterizations of estimator performance for other graph classes (e.g., trees) and for the general case. These characterizations are used to determine the role of measurement and actuation locations in estimation performance. Finally, application of our results in dynamical‐network security is illustrated through a simple example, and a concrete procedure for network mode estimation that draws on our structural results is introduced to conclude the article. Copyright © 2014 John Wiley & Sons, Ltd.     

基于UKF的增长型模糊神经网络设计

为了实现模糊神经网络结构和参数的同时调整,提出一种基于无迹卡尔曼滤波(UKF)的增长型模糊神经网络(UKF-GFNN).首先,利用UKF对模糊神经网络的参数进行调整;然后,设计一种基于隐含层神经元输出强度的模糊规则增长机制,实现模糊神经网络的结构增长;最后,将所提出的增长型模糊神经网络应用于非线性系统建模.实验结果显示,基于UKF的增长型模糊神经网络能够实现结构和参数的自校正,并且具有较高的建模精度.     

COMBINING A RADIAL BASIS NEURAL NETWORK WITH TIME SERIES ANALYSIS TECHNIQUES TO PREDICT MANUFACTURING PROCESS PARAMETERS

The accurate prediction of the values of critical quality parameters of a product during the production stage is a key factor in the success of a manufacturing operation. Neural network algorithms have been used to successfully predict process parameter values. However, techniques to further improve the predictive capability of neural network models are sought. Thus, an analysis was conducted to determine if the predictive capability of the network would he improved if the prediction from a time series model of a manufacturing process parameter were included in the training data set of a radial basis function neural network model. A manufacturing process data set was evaluated, and the use of the time series model prediction significantly improved the neural network's prediction of critical process parameters. Often in a manufacturing environment, the collection of adequate amounts of data for network training is difficult. This integrated technique offers potential for improving network performance without collecting additional data.     

Artificial neural networks applied to odor classification for chemical compounds

Three-layer artificial neural networks (ANN) with back-propagation of error were used to classify the odors of chemical compounds. The network's architecture and parameters were optimized, and an empirical rule for dynamically adjusting the learning rate of network was put forward. It was found that the network gave the maximum recognition rate of 100% and converged quickly by using the plural semiconductor gas sensors' response data in combination with the molecular structure codes of odorants. The network's predictive ability for untrained samples was also satisfactory.     

提高小样本高光谱图像分类性能的变维卷积神经网络

目的 为了解决基于卷积神经网络的算法对高光谱图像小样本分类精度较低、模型结构复杂和计算量大的问题,提出了一种变维卷积神经网络。方法 变维卷积神经网络对高光谱分类过程可根据内部特征图维度的变化分为空—谱信息融合、降维、混合特征提取与空—谱联合分类的过程。这种变维结构通过改变特征映射的维度,简化了网络结构并减少了计算量,并通过对空—谱信息的充分提取提高了卷积神经网络对小样本高光谱图像分类的精度。结果 实验分为变维卷积神经网络的性能分析实验与分类性能对比实验,所用的数据集为Indian Pines和Pavia University Scene数据集。通过实验可知,变维卷积神经网络对高光谱小样本可取得较高的分类精度,在Indian Pines和Pavia University Scene数据集上的总体分类精度分别为87.87%和98.18%,与其他分类算法对比有较明显的性能优势。结论 实验结果表明,合理的参数优化可有效提高变维卷积神经网络的分类精度,这种变维模型可较大程度提高对高光谱图像中小样本数据的分类性能,并可进一步推广到其他与高光谱图像相关的深度学习分类模型中。     

基于Kalman滤波和神经网络的MEMS陀螺温度漂移补偿

MEMS陀螺温度漂移严重影响系统的测量精度。传统的BP神经网络建模补偿容易使权值和阈值陷入局部极小值,导致网络训练失败。陀螺输出信号中的高频噪声也会影响模型精度。针对上述问题,该文提出一种Kalman滤波结合粒子群算法(PSO)优化BP神经网络的MEMS陀螺温度漂移补偿方法。首先对陀螺进行了温度漂移测试实验,然后采用Kalman滤波对实验数据进行降噪,最后建立陀螺温度漂移模型,从而实现温度漂移的补偿。实验结果表明,采用该方法补偿后MEMS陀螺在不同温度下的输出方差降低了65.09%,与传统的BP神经网络相比补偿精度明显提高。     

GPS动态定位中卡尔曼滤波模型的建立及其强跟踪算法研究

提出一种改进的强跟踪卡尔曼滤波算法,应用于GPS动态定位滤波中获得明显效果。首先建立了一种新的GPS动态定位滤波模型,该模型与以往采用的非线性卡尔曼滤波模型相比,具有模型简单、实时性好的特点。为了进一步提高滤波器的动态性能,改进了文献[1]中的强跟踪滤波算法,大大提高了滤波器的跟踪能力。     

大数据融合模型的智能化网络安全检测方法

针对传统智能化网络安全检测平台处理数据效率低、误差大等问题,文章提出一种新型的解决方案;该方案基于大数据融合模型构建新型的智能化网络安全检测平台,采用卡尔曼滤波算法、采用数据融合分类算法和模糊推理算法3种方法结合构建出数据融合模型来对网络安全检测数据进行运算与处理;其中,采用卡尔曼滤波算法进行改进,对原始网络安全检测数据进行滤波降低噪声干扰,提高数据的精准度;通过SAE稀疏自动编码器自主提取网络安全检测数据的特征信息,之后K-means聚类算法对SAE稀疏自动编码器输出的数据进行处理,通过模糊推理算法调整权值;试验表明,文章所提方案克服了现有技术存在的不足,显著提高了处理数据效率和精准度,在数据量为2 TB的环境下,本研究方法的误差低至6.9％.     

Drought estimation with neural networks

An artificial neural network model is presented to derive streamflow precipitation data. It is tested with actual data coming from a nearby river, referred to a basin area of 356 km² and a time period of 11 years. A feedforward multilayer perception with linear output has been built to deal with this problem. The dynamics are caught by the filter structure of the input layer.

A special study on crossing properties, based on training sample selection,is made to measure the performance of the network for drought analysis. Sample selection leads to increased accuracy within the sample range and degraded performance for points that are clearly out. Predicted number of droughts, average drought length and deficit are compared with the actual data. The results show that very simple neural network models can give fine results.     

基于EKF的多小波神经网络综合模型在变形预测中的应用

为了及时有效地对建筑物的变形进行预测,在对多小波、Kalman滤波与神经网络这三种变形预测和建模的有力工具研究的基础上,将多小波分析、神经网络强有力的逼近能力以及Kalman滤波的迭代计算和最优估计的优点有机地结合起来,建立了一种新的变形预测方法:基于扩展Kalman滤波(简称为EKF)的多小波神经网络变形预测模型;通过变形预测实验表明该方法具有较高的精度,较快的速度,是一种能快速高效精确预测变形体变形的方法。     

一种基于BiLSTM的低速率DDoS攻击检测方法

低速率分布式拒绝服务（Low-rate Distributed Denial of Service, LDDoS）攻击是一种新型的DDoS攻击方式,因其具有低速率、周期性和隐蔽性等特点,可躲避传统的DDoS攻击检测技术,更加难于检测和防御。本文提出一种基于特征选择和双向长短期记忆（Bidirectional Long Short Term Memory, BiLSTM）神经网络结合的LDDoS攻击检测方法。该方法使用分层交叉验证的递归特征消除（Recursive Feature Elimination CV, REFCV）特征选择算法挖掘双向流中最优的11个特征集合作为神经网络的输入,建立基于BiLSTM神经网络模型的LDDoS攻击检测分类器进行分类,达到LDDoS攻击检测的目的。实验结果表明该方法比卡尔曼滤波和NCAS算法有较高的检测率,误报率和漏报率都很低。     

Novel L1 neural network adaptive control architecture with guaranteed transient performance.

In this paper, we present a novel neural network (NN) adaptive control architecture with guaranteed transient performance. With this new architecture, both input and output signals of an uncertain nonlinear system follow a desired linear system during the transient phase, in addition to stable tracking. This new architecture uses a low-pass filter in the feedback loop, which consequently enables to enforce the desired transient performance by increasing the adaptation gain. For the guaranteed transient performance of both input and output signals of the uncertain nonlinear system, the L1 gain of a cascaded system, comprised of the low-pass filter and the closed-loop desired reference model, is required to be less than the inverse of the Lipschitz constant of the unknown nonlinearities in the system. The tools from this paper can be used to develop a theoretically justified verification and validation framework for NN adaptive controllers. Simulation results illustrate the theoretical findings.     

面向人机对话意图分类的混合神经网络模型

随着人机对话的不断发展,让计算机能够准确地理解用户查询意图,对整个人机对话领域都有着重要意义.意图分类的主要目标是在人机对话的过程中判断用户的意图,提升人机对话系统的准确度与自然度.首先分析多个分类模型在意图分类任务上的优缺点.在此基础上,提出一种混合神经网络模型,综合利用多个深度网络模型的多样性输出.在输入特征预处理上,采用语言模型词向量,将语言模型拥有的语义挖掘能力应用到混合网络中,可以进一步提升模型的表达能力.所提出的混合神经网络模型相对于最好的基准模型在两份数据集上分别取得了2.95%和3.85%的性能提升.新模型在该数据上取得了最优的性能.