A Characterization of Continuous Module Homomorphisms on Random Semi—Normed Modules and Its Applications

In this paper, a characterization of continuous module homomorphisms on random semi-normed modules is first given; then the characterization is further used to show that the Hahn-Banach type of extension theorem is still true for continuous module homomorphisms on random semi-normed modules.

Upper semi-continuity of attractors for multivalued semi-flow under random perturbation

In this paper the upper semi-continuity of global attractors for multivalued semi-flows under random perturbation was studied. First, the existence of random attractors for multivalued random semi-flows was considered, then it was proved that the global attractors for multivalue semi-flows are the upper semi-continuity under random perturbation. This result can be used in the numerical approximation of multivalued semi-flows and non-autonomous perturbation of multivalued semi-flows.

Definition of Laplace Transforms for Distribution of the First Passage of Zero Level of the Semi-Markov Random Process with Positive Tendency and Negative Jump

One of the important problems of stochastic process theory is to define the Laplace transforms for the distribution of semi-markov random processes. With this purpose, we will investigate the semimarkov random processes with positive tendency and negative jump in this article. The first passage of the zero level of the process will be included as a random variable. The Laplace transforms for the distribution of this random variable is defined. The parameters of the distribution will be calculated on the basis of the final results.

Existence and Upper Semi-Continuity of Random Attractors for Nonclassical Diffusion Equation with Multiplicative Noise on Rn

This paper is concerned with the existence and upper semi-continuity of random attractors for the nonclassical diffusion equation with arbitrary polynomial growth nonlinearity and multiplicative noise in H1(Rn). First, we study the existence and uniqueness of solutions by a noise arising in a continuous random dynamical system and the asymptotic compactness is established by using uniform tail estimate technique, and then the existence of random attractors for the nonclassical diffusion equation with arbitrary polynomial growth nonlinearity. As a motivation of our results, we prove an existence and upper semi-continuity of random attractors with respect to the nonlinearity that enters the system together with the noise.

地震多属性储层厚度半定量预测--以鄂尔多斯盆地东南部山2段储层为例

鄂尔多斯盆地东南部山2段致密气储层薄、埋深大、构造平缓、上下围岩存在煤、灰岩等强反射界面,利用低频、弱信号黄土塬二维地震资料,定量预测储层厚度难度大。以延安气田延113-延133井区二叠系山西组山23储层为研究对象,对储层厚度进行分级,基于二维地震资料标准化处理的结果提取地震属性,分析二维地震属性与实钻储层厚度级别之间的关系,建立了多个随机森林算法融合多属性预测方案,进行优化筛选半定量储层厚度预测方案。优选方案预测山23砂体厚度与实钻结果相关系数达0.77。该方法可加深对研究区山23河道展布认识,落实并增加该区地质储量,取得了显著的建井成效。

随机反应扩散方程的随机吸引子的高阶稳定性

本文研究带加性白噪声的随机反应扩散方程的随机吸引子关于噪声强度的稳定性。首先,在非线性函数满足更一般条件下,在初始空间L^(2)(R^(N))中获得随机微分方程的解收敛到确定方程的解,从而得到随机吸引子的上半连续性;然后,利用非线性分解和差分估计,证明L^(p)(R^(N))(p>2)空间中解的收敛性和随机吸引子的上半连续性,其中p是非线性函数的增长指数。

滑坡易发性预测建模的不确定性:不同“非滑坡样本”选择方式的影响

滑坡易发性预测建模中如何选择非滑坡是影响建模结果的重要不确定因素。为研究不同非滑坡选择方式的影响规律,拟用5种方式,即全区随机、坡度低于5°区域、滑坡缓冲300 m外区域、信息量(IV)法、半监督法来选择出与滑坡等比例的非滑坡样本;进一步将各选择方式与随机森林(RF)耦合构建随机RF、低坡度RF、缓冲区RF、IV–RF及半监督RF等模型。以江西南康区为例,获取高程、岩性、公路密度等19种环境因子和233个滑坡编录,将滑坡编录划分为2598个滑坡栅格单元构建上述耦合模型的输入–输出数据集。再采用预测精度和易发性指数分布等指标分析其建模不确定性。进一步针对耦合模型预测的滑坡易发性指数分布不合理等问题,在半监督RF建模时采用滑坡与非滑坡比例为1∶2的样本集开展建模并与1∶1等比例样本集工况作对比。结果表明:1)低坡度RF、缓冲区RF、IV–RF和半监督RF等模型的预测精度均大幅优于随机RF模型,可见准确选择非滑坡样本对易发性建模至关重要;2)半监督RF模型选择非滑坡样本的建模性能最优,且半监督RF在滑坡∶非滑坡=1∶2比其在1∶1时预测的易发性指数分布规律更准确可信。后续研究中有必要更深入探索滑坡与非滑坡样本的比例问题。

基于闭合频繁模式的半随机森林数据流分类算法

提出了一种基于闭合频繁模式的半随机森林数据流分类算法(Semi-Random Forest based on Closed Frequent Pattern,SRFCFP),以解决数据流中噪声和概念漂移问题。SRFCFP利用闭合频繁模式对数据流进行表示,去除冗余信息和噪声,突出数据特征。采用半随机森林建立分类模型,并通过基于时间衰减的模式集更新机制适应数据流的无限性。为了检测概念漂移并及时适应,引入了一种模式集差异性度量方式,用于测量数据分布变化。实验结果表明,在MOA平台下使用真实和合成数据集,SRFCFP在平均精度上超越了相关对比算法,并能有效处理数据流中的概念漂移和噪声问题。

CRF机制结合LDA的病历文书后结构化系统的应用

为了提高病历文书结构化分类准确度,提出利用条件随机场(CRF)半监督词典分词算法结合隐式狄利克雷分布(LDA)医学病历文本分类算法,构建出CRF机制结合LDA的病历文书后结构化系统。结果表明:当主题数量为40时,LDA主题建模的困惑度最小值为-6.97,与初始困惑度相比,LDA主题建模困惑度下降9.76%;当主题数量为3时,得到了一致性值的最低值为0.361;当主题数量为40时,得到了一致性值的最大值为0.442,与最低值相比,LDA主题建模一致性值上升22.44%。综上可以看出,研究的CRF机制结合LDA的病历文书后结构化系统具有较好的应用效果。

基于数据挖掘的提高汽车座舱推送类产品触发成功率的研究

随着汽车行业向智能化和网联化转型,智能座舱体验已成为中高端车型的核心竞争力。传统的推送功能研发方法受限于人工配置的随意性,导致触发成功率低、触达精准性差等问题,为了解决这些问题,提出了一种基于车联网数据的智能座舱场景推送模型。该模型通过深入分析功能数据特征,运用监督学习技术训练多个算法,采用半监督学习方法克服数据标签不足的挑战,利用终端数据埋点技术,对模型进行精细优化,以提高推送的准确性和效率。实验结果表明,与传统方法相比,该模型在触发成功率和触达精准性上取得了显著提升。这一成果不仅为智能座舱技术的发展提供了有力支持,也为用户带来了更加个性化、高品质的座舱体验。

基于生成式对抗网络和改进区域建议网络的输电线路杆塔缺陷检测方法

为了减少输电线路杆塔缺陷检测过程中受噪声信号和装置性能等因素的干扰,提高输电线路杆塔缺陷检测的正确率和检测效率。提出一种基于生成式对抗网络(generative adversarial networks,GAN)和改进区域建议网络(region proposal network,RPN)的输电线路杆塔缺陷检测方法。采用GAN采集输电线路杆塔的显著性图像,并利用半软阈值函数模型剔除图像中的噪声,避免噪声对缺陷检测过程产生影响。通过随机森林决策树提取输电线路杆塔图像的轮廓特征,基于多尺度算法对RPN进行改进,将特征输入到改进RPN模型中,通过缺陷的定位、分割完成输电线路杆塔的缺陷检测。试验结果表明,所提方法的输电线路杆塔缺陷检测正确率较高,具有较好的缺陷检测效果和检测效率,从而有利于提高输电线路杆塔缺陷检测的质量,减少电力事故的出现。

Stationary Distribution of Random Motion with Delay in Reflecting Boundaries

In this paper we study a continuous time random walk in the line with two boundaries [a,b], a < b. The particle can move in any of two directions with different velocities v1 and v2. We consider a special type of boundary which can trap the particle for a random time. We found closed-form expressions for the stationary distribution of the position of the particle not only for the alternating Markov process but also for a broad class of semi-Markov processes.

Semi-order Fuzzy Supermartingales and Submartingales

There are some mathematical models(see Example2.4)and analogous results in standard martingale theorywhich can not be described by the usual fuzzy martingaletheory because of the lack of corresponding semi-orderin the fuzzy number space(E^n,D).In this paper,asuitable semi-order in the fuzzy number space(E^n,D)and the semi-order fuzzy supermartingale and submar-tingale are introduced,the charaterlstics of semi-ordersupermartingales and submartingales,as well as theDood’s stopping theorem for them(the bounded stoppingtimes theorem and the general stopping times theoremfor a class of closable semi-order fuzzy supermartin-gales and submartingales)are established.

基于随机森林与半监督学习的钢箱梁桥模块化智能选型研究

为指导中等跨径钢箱梁桥智能选型设计,将钢桥模块化理论与随机森林算法相结合用于初步选型,采用半监督学习的方法增强数据集,提出一种中等跨径钢箱梁桥模块化智能选型方法。该方法首先选择桥梁10个特征作为自变量,将钢箱梁的9种横截面形式作为因变量,收集75条专家评价数据作为训练集;其次通过算法适配和对比分析,采用遗传-随机森林的组合算法对结果的分类精度相较于原始的随机森林算法模型准确率提升了约10%;然后采用半监督学习算法对训练数据集进行增强,在不改变算法的同时进一步提高了模型准确率,整体模型准确率接近0.9;最后基于上述算法和前后端技术实现在Web平台上的快速部署和三维结果展示。

Modal Correlation Based Analysis on Dynamic Characters of a Commercial Semi-Trailer's Chassis Frames

In order to evaluate two different schemes＇ structural dynamic characters, dynamic response analysis of a commercial truck＇s main chassis frames is carried out. On the basis of correlation study between the tested and calculated modal results, the assembled frames＇ finite element analysis （FEA） models with sufficient precision are built up. Random response analysis in frequency domain is carried out with these FEA models, RMS values of yon Mises and main principle stresses of these two frames are obtained. It shows that the analysis resuits of the distributing tendency and concrete value ranges are coincident very well with test results. And from the results, it could be concluded that frames of scheme A endures relative better loading conditions and should be adopted as the final scheme.

Numerical Study on the Probability of Fatigue Failure of a Semi-Submerged Platform Structure

On the basis of Miner's linear cumulative damage theorem, random variables are introduced to evaluate objectively the fatigue damage of a semi-submerged platform structure and a method is presented to analyze the fatigue reliability of the structure in its design life. The reliability of the method is verified through numerical examples and some conclusions are drawn, which have certain guiding and reference value for design and inspection.

基于与测线交切半迹长的岩体随机结构面平均直径估算

假定岩体随机结构面形状为薄圆盘,为能科学合理估算其平均直径,基于半迹长测线法,在未对分布形式作任何假设的条件下,通过理论推导获得与测线交切半迹长和岩体随机结构面直径之间的普适关系式.为便于工程应用,假设岩体随机结构面直径服从负指数分布,推导出其平均直径和与测线交切平均半迹长的关系式,基于该式提出一种岩体随机结构面平均直径的估算方法.采用岩体随机结构面三维网络模拟技术对此方法进行应用检验,结果表明,该方法具有较高的准确性与较强的工程适用性.

基于半监督网络表示学习方法的基因功能预测

直接定义在网络邻接矩阵上运行的特征提取方法存在着结构信息提取不充分的问题。基于此,提出一种融合多网络表示学习特征的基因功能预测方法。利用随机游走模型捕捉基因网络的拓扑结构特征,基于此特征使用一个半监督自编码器学习基因节点的低维表示。应用卷积神经网络融合多网络基因节点的低维表示信息预测基因功能。实验结果表明,在人类和酵母菌数据集上该方法能够提高基因功能预测的效果。

基于半监督的3D肝脏CT自动分割方法研究

肝癌是常见癌症之一,有着较高的死亡率,精准分割肝癌区域是辅助诊断治疗的重要前提。然而肝脏CT图像需要专业的医师进行标注,有标签数据较少,获取途径单一。针对分割腹部肝脏CT图像需要大量高质量标签并且较难获取的问题,提出了一种采用协同训练的半监督学习分割方法。首先,将有标签数据输入3D U-Net和3D Res U-Net进行有监督训练,保存训练得到的两个分割模型,在两个分割模型中分别对无标签数据进行预测;然后,挑选预测的结果,再加入全连接条件随机场处理挑选出的伪标签,细化伪标签的边缘信息,提升伪标签的精确度;最后,加入到训练集中,重复上述步骤直到分割结果的Dice相似系数停止提升时结束训练。实验在LiTS2017 Challenge肝脏数据集上进行测试,结果表明,在有标签数据集占总数据集的30%时,该方法的Dice值达到90.22%,几乎与全监督3D Res U-Net分割结果持平,说明该半监督学习方法是有效的。