基于多层神经网络的智能家居入侵检测方法

依托物联网技术的智能家居面临多重信息安全风险,现有智能家居入侵检测方案存在难以处理大量高维度数据、检测率低、误检率高、依赖经验确定网络层数等问题。提出一种融合深度学习与模糊神经网络的多层神经网络入侵检测方法;基于深度学习完成数据特征的学习,将高维数据映射为低维数据;基于网络重构误差训练并优化确定网络深度。仿真测试结果表明,该方案可有效提高对攻击行为的检测准确率和检测效率;针对远程非法访问的检测率可达到94%,对拒绝服务攻击的检测准确率可达96%,对网络中新型攻击的检测率超过60%。     

基于无监督迁移成分分析和深度信念网络的轴承故障诊断方法

针对轴承故障样本少导致识别精度低的问题,提出一种基于无监督迁移成分分析(unsupervised transfer component analysis,UTCA)和深度信念网络(deep belief network,DBN)的故障诊断方法。首先利用UTCA的核函数将不同工况样本特征映射到一个共享再生核Hilbert空间中,使得源域和目标域样本集更加相似,并通过最大均值偏差嵌入法(maximum mean discrepancy embedding,MMDE)判断能够迁移的源域数据,将源域样本迁移到目标域中,为深度学习提供充足的训练样本,解决了实际故障样本较少的问题;然后采用DBN模型对源域样本进行训练,再对映射后无标记的目标域样本进行故障诊断分析。利用不同工况下的滚动轴承实验数据进行算法验证,结果表明,与普通DBN、SVM、BPNN以及传统机器学习-UTCA融合方法相比,本文方法对滚动轴承故障的诊断精度更高。     

基于深度置信网络和信息融合技术的轴承故障诊断

提出一种基于深度置信网络(DBN)和信息融合技术的轴承故障诊断新方法。首先采用集合经验模式分解将轴承振动时域信号分解为若干个固有模态函数,并分别输入至若干个DBN中进行故障状态识别,然后通过简单投票法将每个DBN识别的结果进行决策层信息融合,从而得到轴承故障的最终诊断结果。通过对单负载和多负载下不同类型和不同损伤程度的滚动轴承故障诊断进行实例分析,验证了本文方法的有效性和精确性。     

Micro-level reserving for general insurance claims using a long short-term memory network

Detailed information about individual claims are completely ignored when insurance claims data are aggregated and structured in development triangles for loss reserving. In the hope of extracting predictive power from the individual claims characteristics, researchers have recently proposed to use micro-level loss reserving approaches. We introduce a discrete-time individual reserving framework incorporating granular information in a deep learning approach named Long Short-Term Memory (LSTM) neural network. At each time period, the network has two tasks: first, classifying whether there is a payment or a recovery, and second, predicting the corresponding non-zero amount, if any. Based on a generalized Pareto model for excess payments over a threshold, we adjust the LSTM reserve prediction to account for extreme payments. We illustrate the estimation procedure on a simulated and a real general insurance dataset. We compare our approach with the chain-ladder aggregate method using the predictive outstanding loss estimates and their actual values.     

Unbound States in an Almost Infinite Deep Potential

The behavior of a quantum particle in an almost infinite deep potential is investigated. By means of analytical and numerical methods, it is demonstrated that even if the potential is infinitely deep, the particle's eigenstate can still be unbound, which contradicts the previous belief that an infinite deep potential can only sustain bound states. These findings shed new light on the eigenstate problem of quantum mechanics.     

基于光谱图像的森林病虫害自动检测方法

我国天然林区分布范围广,地形复杂,依靠传统的护林员巡检方式进行林木病虫害防治,效率较低,难于及时发现早期的林木病虫害,可能因此错过防治的最佳时机。针对该问题,设计了一种基于多光谱图像检测林木病虫害的深度学习网络,研发了一套检测软件,通过无人机挂飞实验,利用搭建的深度学习网络,完成林区染病区检测,对检测结果进行了分析。     

利用深度神经网络实现分布式相干瑞利光纤振动事件分类

该文利用分布式相干瑞利光纤传感系统,在西气东输一线无锡至苏州段开展现场测试,采集了光纤沿线车辆行走、机械挖掘、人工锄地、定向钻孔等8种振动作业产生的光纤信号,并提出了一种具有5层结构的全连接深度神经网络用于振动事件分类识别以实现不同振动作业的分级管理。振动作业产生的光纤信号能量集中在低频,该文利用梅尔对数频率的非均匀特性提取了25维单帧信号特征量,并将连续40帧信号特征量组合成高维向量作为网络输入特征向量,实现对不同振动作业时变特性的建模。分类识别结果表明,基于深度神经网络结构的振动信号分类识别器能够有效识别不同振动作业类型,实际线路实验验证了该文算法的有效性。     

基于LSSVM和RFID的智能门禁身份认证系统设计

针对目前已有的基于RFID的身份识别系统,往往仅从系统角度来设计系统,没有和计算机视觉技术结合,从而无法进一步提高安全性能。为了解决此问题,文中设计了一种基于LSSVM(Least Square Support Vector Machine, LSSVM)和RFID(Radio Frequency Identification, RFID)的智能门禁身份认证系统。首先,设计了系统总体框架并描述了身份认证原理。然后,对系统的RFID子系统,微处理器和接口均进行了详细设计和描述,在此基础上,将计算机视觉技术加入到识别过程中,采用多层RBM组成的深度神经网络对人脸进行自动特征提取,通过比较差异算法训练深度神经网络,最后,根据RFID标签和提出的人脸特征,采用多个LSSVM进行人脸识别,并将投票结果作为最终识别结果。通过部署实验进行测试,结果证明文中设计的系统能实时有效地对智能门禁进行身份认证,与其它方法相比,具有识别率高和识别效率高的优点,具有较大的优越性。     

Structural characterization of the aquaporin inhibitor 2‐nicotinamido‐1,3,4‐thiadiazole

Nicotinamides are a class of compounds with a wide variety of applications, from use as antimicrobial agents to inhibitors of biological processes. These compounds are also cofactors, which are necessary components of metabolic processes. Structural modification gives rise to the activities observed. Similarly, 1,3,4‐thiadiazoles have been shown to possess antioxidant, antimicrobial, or anti‐inflammatory biological activity. To take advantage of each of the inherent characteristics of the two aforementioned functional groups, 2‐nicotinamido‐1,3,4‐thiadiazole, C₈H₆N₄OS, was synthesized. Since defining chemical connectivity is paramount in understanding biological activity, in this report, the structural characterization of 2‐nicotinamido‐1,3,4‐thiadiazole has been carried out using X‐ray crystallographic methods. The NMR‐derived assignments were made possible by utilizing one‐ (1D) and two‐dimensional (2D) NMR techniques. In addition, UV–Visible and IR spectroscopies, and elemental analysis were used to fully characterize the product synthesized by the one‐step reaction between nicotinoyl chloride hydrochloride and 2‐amino‐1,3,4‐thiadiazole. Computational parameters related to blood–brain barrier permeability are also presented.     

Large-Scale Turbulent Vortical Structures inside a Sudden Expansion Cylinder Chamber

A large eddy simulation (LES) is performed for turbulent flow around a bluff body inside a sudden expansion cylinder chamber, a configuration which resembles a premixed gas turbine combustor. To promote turbulent mixing and to accommodate flame stability, a flame holder is installed inside the combustion chamber. The Smagorinsky model and the Lagrangian dynamic subgrid-scale model are employed and tested. The calculated Reynolds number is 5,000 based on the bulk velocity and the diameter of inlet pipe. The simulation code is constructed by using a general coordinate system based on the physical contravariant velocity components. The predicted turbulent statistics are evaluated by comparing with the laser-doppler velocimetry (LDV) measurement data. The agreement of LES with the experimental data is shown to be satisfactory. Emphasis is placed on the time-dependent evolutions of turbulent vortical structures behind the flame holder. The numerical flow visualizations depict the behavior of large-scale vortices. The turbulent behavior behind the flame holder is analyzed by visualizing the sectional views of vortical structure. This revised version was published online in July 2006 with corrections to the Cover Date.