Temporal evolution of the focal mechanism consistency of the 2021 Yangbi M_(S) 6.4 earthquake sequence in Yunnan

Using the Cut And Paste(CAP)method,we invert the focal mechanism of 38 moderate earthquakes(M_(S)≥3.0)recorded by Yunnan seismic network and analyze the corresponding focal mechanism consistency based on the minimum spatial rotation angle.Our results indicate that the M_(S)6.4 mainshock is induced by a lateral strike slip fault(with a rake angle of~-165°)and a little normal-faulting component event along a nearly vertical plane(dipping angle~79° and strike~138°).Combining our results with high resolution catalog,we argue that the seismogenic fault of this earthquake sequence is a secondary fault western to the major Weixi-Qiaohou-Weishan fault.The focal mechanism evolution can be divided into three periods.During the first period,the foreshock sequence,the focal mechanism consistency is the highest(KA<36°);during the second period which is shortly after the mainshock,the focal mechanism shows strong variation with KA ranging from 8° to 110°;during the third period,the seismicity becomes weak and the focal mechanism of the earthquakes becomes more consistent than the second period(18°<KA<73°).We suggest that the KA,to some extent,represents the coherence between local tectonic stress regime and the stress state of each individual earthquake.Furthermore,high focal mechanism consistency and high linearity of seismic distribution may serve as indicators for the identification of foreshock sequence.

Consistency analysis of accelerated degradation mechanism based on gray theory

A fundamental premise of an accelerated testing is that the failure mechanism under elevated and normal stress levels should remain the same. Thus, verification of the consistency of failure mechanisms is essential during an accelerated testing. A new consistency analysis method based on the gray theory is pro- posed for complex products. First of all, existing consistency ana- lysis methods are reviewed with a focus on the comparison of the differences among them. Then, the proposed consistency ana- lysis method is introduced. Two effective gray prediction models, gray dynamic model and new information and equal dimensional （NIED） model, are adapted in the proposed method. The process to determine the dimension of NIED model is also discussed, and a decision rule is expanded. Based on that, the procedure of ap- plying the new consistent analysis method is developed. Finally, a case study of the consistency analysis of a reliability enhancement testing is conducted to demonstrate and validate the proposed method.

Types of focal mechanism solutions and parameter consistency of the sub-blocks in Sichuan and Yunnan Provinces

Based on P- and S-wave amplitudes and some clear initial P-wave motion data, we calculated focal mechanism solutions of 928 M≥2.5 earthquakes （1994-2005） in four sub-blocks of Sichuan and Yunnan Provinces, namely Sichuan-Qinghai, Yajiang, Central Sichuan and Central Yunnan blocks. Combining these calculation results with those of the focal mechanism solutions of moderately strong earthquakes, we analyzed the stress field characteristics and dislocation types of seismogenic faults that are distributed in the four sub-blocks. The orientation of principal compressive stress for each block is： EW in Sichuan-Qinghai, ESE or SE in Yajiang, Central Sichuan and Central Yunnan blocks. Based on a great deal of focal mechanism data, we designed a program and calculated the directions of the principal stress tensors, σ1, σ2 and σ3, for the four blocks. Meanwhile, we estimated the difference （also referred to as consistency parameter θ^- ） between the force axis direction of focal mechanism solution and the direction of the mean stress tensor of each block. Then we further analyzed the variation of θ^- versus time and the dislocation types of seismogenic faults. Through determination of focal mechanism solutions for each block, we present information on the variation in θ^- value and dislocation types of seismogenic faults.

Consistent Distribution of Stress before Strong Earthquake from Focal Mechanisms

At some stage of a strong earthquake preparation, the focal mechanisms of small earthquakes have roughly the same direction with the tectonic stress field. According to this feature, we define the angle between P, B and T axis of focal mechanisms and the three stress axes of tectonic stress field as the consistency parameter a in studying the dynamic changes of stress fields in earthquake preparation areas. We mainly analyze the changes of the consistency parameter a of the Mw8. 3 Knril island arc earthquake and the Mw8. 4 Peru earthquake. Our study shows that before the strong earthquakes, the earthquake area saw a low consistency, and the focal mechanisms of a series of small earthquakes had small differences in the directions with the tectonic stress field, which means the foreshocks were under the control of the stress field. On the other hand, a higher consistency means the focal mechanisms of their aftershocks are scattered and have big differences in the directions with the tectonic field, which indicate that the control of background stress field starts weakening.

Difference between focal mechanisms of Dayao earthquake doublet sequence

Using the maximum amplitude ratios of vertical component of P and S waves recorded by a regional network, 921 focal mechanisms of Dayao earthquake doublet sequence are determined by means of synthetic seismograms of a point source of dislocation in a plane layered medium. Among them, 389 focal mechanisms are in the aftershock sequence of M6.2 earthquake occurred on 21 July, 2003 and the other 532 focal mechanisms are in the aftershock sequence of M6.1 earthquake occurred on 16 October, 2003 in Dayao, Yurman. The focal mechanism consistent parameter a of the two aftershock sequences are calculated and analyzed. According to the focal mechanism consistent parameter a, the focal mechanisms of the first aftershock sequence are more consistent than those of the second. According to the comparison of CMT solutions of the two M6 earthquakes, the physical mechanism of the doublet in the intra-plate earthquake is very complex, and many processes are involved and interacted with each other. This doublet provides insights into earthquake clustering, triggering and stress cycling.

Thermodynamic Consistency of Plate and Shell Mathematical Models in the Context of Classical and Non-Classical Continuum Mechanics and a Thermodynamically Consistent New Thermoelastic Formulation

Inclusion of dissipation and memory mechanisms, non-classical elasticity and thermal effects in the currently used plate/shell mathematical models require that we establish if these mathematical models can be derived using the conservation and balance laws of continuum mechanics in conjunction with the corresponding kinematic assumptions. This is referred to as thermodynamic consistency of the mathematical models. Thermodynamic consistency ensures thermodynamic equilibrium during the evolution of the deformation. When the mathematical models are thermodynamically consistent, the second law of thermodynamics facilitates consistent derivations of constitutive theories in the presence of dissipation and memory mechanisms. This is the main motivation for the work presented in this paper. In the currently used mathematical models for plates/shells based on the assumed kinematic relations, energy functional is constructed over the volume consisting of kinetic energy, strain energy and the potential energy of the loads. The Euler’s equations derived from the first variation of the energy functional for arbitrary length when set to zero yield the mathematical model(s) for the deforming plates/shells. Alternatively, principle of virtual work can also be used to derive the same mathematical model(s). For linear elastic reversible deformation physics with small deformation and small strain, these two approaches, based on energy functional and the principle of virtual work, yield the same mathematical models. These mathematical models hold for reversible mechanical deformation. In this paper, we examine whether the currently used plate/shell mathematical models with the corresponding kinematic assumptions can be derived using the conservation and balance laws of classical or non-classical continuum mechanics. The mathematical models based on Kirchhoff hypothesis (classical plate theory, CPT) and first order shear deformation theory (FSDT) that are representative of most mathematical models for plates/shells are investigated in this paper for their thermodynamic consistency. This is followed by the details of a general and higher order thermodynamically consistent plate/shell thermoelastic mathematical model that is free of a priori consideration of kinematic assumptions and remains valid for very thin as well as thick plates/shells with comprehensive nonlinear constitutive theories based on integrity. Model problem studies are presented for small deformation behavior of linear elastic plates in the absence of thermal effects and the results are compared with CPT and FSDT mathematical models.

Personalized Dialogue Generation Model Based on BERT and Hierarchical Copy Mechanism

Despite the great advances in generative dialogue systems, existing dialogue generation models are still unsatisfactory in maintaining persona consistency. In order to make the dialogue generation model generate more persona-consistent responses, this paper proposes a model named BERT-HCM (Personalized Dialogue Generation Model Based on BERT and Hierarchical Copy Mechanism). The model uses an encoder based on BERT initialization to encode persona information and dialogue queries and subsequently uses a Transformer decoder incorporating a hierarchical copy mechanism to dynamically copy the input-side content to guide the model in generating responses. The experimental results show that the proposed model improves on both automatic and human evaluation metrics compared to the baseline model and is able to generate more fluent, relevant and persona-consistent responses.

具有细晶/混晶交替分布组织特征的AZ31镁合金轧板拉伸变形行为仿真

通过中等应变速率轧制获取细晶区和混晶区交替分布的AZ31镁合金轧制板材,构建不同组织特征的黏塑性自洽(VPSC)模型,预测板材在室温拉伸下的变形机制、织构演变和力学响应,研究细晶区与混晶区对塑性变形的贡献。结果表明:经410℃、平均应变速率1.9 s^(-1)、4.8 s^(-1)和7.8 s^(-1)制备的轧制板材中细晶区与混晶区呈现板织构特征,细晶区织构组分相较于混晶区更接近ND方向;随着轧制平均应变速率的增加,板材的细晶区与混晶区的织构组分差异逐渐减小。在室温拉伸塑性变形初期阶段,轧制板材的主要变形机制是基面滑移,次要变形机制是棱柱面滑移和锥面滑移,而拉伸孪生几乎不开启。压缩孪生启动时,部分晶粒向TD方向偏转56°,形成TD方向的织构组分。棱柱面滑移活性的降低,有利于增强锥面滑移的分离效应,促使ND方向织构极密度降低和双峰织构形成。轧制板材在拉伸过程中,由于细晶区的织构组分相较于混晶区更接近ND方向,使得在拉伸过程中细晶区相较于混晶区的基面滑移开启较少,非基面滑移开启较多,促使拉伸变形的过程中细晶区起强化作用,混晶区起弱化作用。织构组分差异逐渐减小,使得细晶区与混晶区的变形机制活性趋于一致,促使其在拉伸过程中对抗拉强度贡献也逐渐趋于一致。

基于包络学习和分级结构一致性机制的不平衡集成算法

集成方法是不平衡学习方法的重要分支,然而,现有不平衡集成方法均作用于原样本而没考虑样本的结构信息,因此其效能仍然有限.样本的结构信息包括局部和全局结构信息.为了解决上述问题,本文提出了一种基于深度样本包络网络(Deep Instance Envelope Network,DIEN)和分级结构一致性机制(Hierarchical Structure Consistency Mechanism,HSCM)的不平衡集成学习算法.该算法在考虑局部流形和全局结构信息的情况下,通过多层样本聚类,生成高质量的多层包络样本,从而实现类平衡化.首先,算法基于样本近邻拼接和模糊C均值聚类算法,设计DIEN来挖掘样本的结构信息,得到深度包络样本.然后,设计局部流形结构度量和全局结构分布度量来构建HSCM用于增强层间样本的分布一致性.接着,将DIEN和HSCM结合起来,构建出优化后的深度样本包络网络——DH(DIEN with HSCM).之后,将基分类器应用于包络样本.最后,设计bagging集成学习机制来融合基分类器的预测结果.文末组织了多组实验,采用了十多个公共数据集和有代表性的相关算法进行验证比较.实验结果表明,本文算法在AUC(Area Under Curve),F-measure等四个性能指标上显著最优.

湿基α-半水石膏制备高强石膏制品研究

为实现湿基α-半水石膏制备高强石膏制品,研究了湿基α-半水石膏陈化时间、粉磨时间、浇筑料浆温度及缓凝剂掺量等工艺参数对高强石膏凝结时间和强度的影响,并确定最佳制备工艺。此外,采用X射线衍射、扫描电子显微镜等微观测试手段,揭示了各因素对强度的影响机理。结果表明:陈化时间越长,石膏凝结时间越短,制品强度越低;随着粉磨时间延长,石膏凝结时间显著缩短,制品强度先升高后降低;料浆温度越高,石膏凝结时间越长,制品强度越低。最佳制备工艺:陈化时间5 min、粉磨时间20 s、料浆温度20℃。在此基础上加入0.08%(质量分数)缓凝剂能够有效地延长石膏凝结时间并提高硬化强度,制备的石膏制品性能优异,绝干抗折强度为14.1 MPa,绝干抗压强度为58.3 MPa。微观测试结果表明,各工艺参数会通过改变石膏标稠用水量和水化进程来影响石膏内部结构的密实程度和二水石膏晶体的生长发育,最终影响制品硬化强度。

漾濞6.4级地震前后震源机制一致性时空演化特征

利用CAP方法反演2021年漾濞6.4级地震震源机制,并基于云南地区2015~2022年M≥3.0地震震源机制解结果,采用叠加应力场反演方法反演漾濞6.4级地震前后不同时段的应力张量方差空间分布特征;采用云南地区2013~2022年M≥3.0地震震源机制解结果分析震源区附近震源机制一致性参数的时序变化特征。结果表明:漾濞6.4级地震性质为右旋走滑型,矩震级M_(W)6.03,矩心深度5.8 km,节面Ⅰ走向39°、倾角75°,滑动角-16°;节面Ⅱ走向133°、倾角75°、滑动角-164°。应力张量方差空间演化特征显示,漾濞6.4级地震震源区附近应力张量方差经历了一个低值-显著升高-震前下降形成新低值-震后显著升高的变化过程。应力张量方差时间演化特征显示,漾濞6.4级地震震源区附近应力张量方差在地震前1~2 a达到最高值,后持续下降,下降至最低值后发生转折并趋势回升,整体呈正“V”字型,地震发生在正“V”字型转折后趋势回升阶段。应力张量方差的时空演化特征具有一致性。

基于判别性特征增强的小样本细粒度图像识别

小样本细粒度图像识别是深度学习领域中一个热门的研究课题,其基本任务是在学习有限数量样本的情况下识别出某一大类下的子类别的图像。得益于卷积神经网络的快速发展,小样本细粒度图像识别在精度方面取得了显著的成果,但其性能仍受限于同一子类图像间的高方差以及不同分类任务中判别性特征的差异性。针对上述问题,提出了一种基于判别性特征增强的小样本细粒度图像识别算法(DFENet)。DFENet设计了对称注意力模块来增强类内视觉一致性学习,从而减少背景的影响,提高同类样本之间共享的特征表示的权重。此外,DFENet引入通道维度的判别性特征增强模块,利用支持集样本中同类样本内和不同类样本之间的通道关系进一步挖掘适合于当前任务的判别性特征,以提高识别准确率。在三个经典的细粒度数据集CUB-200-2011,Stanford Dogs,Stanford Cars上进行了广泛的实验。实验结果表明,该方法均取得了有竞争性的结果。

结合构造应力场探究地磁转换函数异常

变化的磁场是联系区域应力场和孕震区地球物理环境信息的天然媒介,而地磁短周期转换函数可以反映地下电阻率等物理信息,本研究利用有界影响估计法计算喀什地磁台地磁短周期转换函数实部Ar绝对值,并使用南天山西段及西昆仑地区的震源机制解资料反演了研究区震源机制一致性参数,其时序变化代表研究区应力场的变化。结合研究区应力场变化尝试探究地磁转换函数实部Ar绝对值异常对构造应力作用的响应。结果显示,当应力作用增强后,地磁转换函数在时间上出现同步或滞后的异常变化,表明应力在不断积累增强后,地下介质电阻率发生变化,从而引起感应磁场变化,因此得出应力变化是造成局部磁异常变化的影响因素之一,验证了地震前感应磁效应现象。通过研究8次中强震前应力场与地磁场转换函数变化,发现在6次中强震前均出现震前异常变化,两者在震前的变化有较好的一致性。研究还发现喀什地磁台地磁转换函数不同周期的实部Ar绝对值对不同方位的地震响应不同,可能与震源区地下介质各向异性差异有关。

高动态范围图像光照学习算法

为解决三维场景光照估计易受天气条件和拍摄视角等多方面因素影响的问题,提出一种基于深度学习的球谐系数估计方法,包括特征提取模块和光照估计模块。使用输入的图像对及预先计算好的光照参数训练网络,特征提取模块提取输入图像的特征,迭代改变球谐系数损失和渲染损失来优化网络,通过光照估计模块预测表达场景光照的球谐系数。实验通过与其它6种主流方法以及3种网络模型进行比较,所用评价指标均有显著性降低,实验结果表明,提出方法可有效恢复光照效果,定量定性地验证了恢复的光照在视觉上是真实的。

面向过程文本的合规性检查方法

合规性检查是过程挖掘领域的重要场景之一,其目标是判断实际运行的业务行为与理想的业务行为是否一致,进而为业务过程管理提供决策依据.传统的合规性检查方法存在度量指标过多、效率低等问题.此外,现有研究在检查过程文本与过程模型之间的合规性时严重依赖专家知识.为此,提出面向过程文本的合规性检查方法.首先,基于过程模型的执行语义生成图轨迹,并利用词向量模型提取图轨迹中的结构特征.同时,引入霍夫曼树提升词向量模型的效率.接着,对过程文本和模型中的活动特征进行提取,并利用孪生机制提升训练效率.最后,对所有特征进行融合,并利用全连接层预测过程文本与过程模型之间的一致性得分.实验表明,所提方法的平均绝对误差值要比已有方法低2个百分点.

结合信誉跳跃一致性哈希的区块链分片协议

区块链面临扩展性问题,目前主要的解决方法是采用分片技术提升交易的处理能力,但分片过后单个分片内节点数量降低,容易造成分片内中心化与拜占庭节点聚集,对网络的安全性产生危害。针对此问题,提出了一个结合信誉跳跃一致性哈希的区块链分片协议。设计了一种节点信誉等级模型,通过节点在网络中的行为赋予节点不同的信誉等级,降低拜占庭节点在网络中的话语权;提出了基于信誉的跳跃一致性哈希分片部分重构算法(reputation-based jump consistent hash sharding partial reconstruction,RJSPR),将信誉值作为节点进行分片重配置的重要考量,减少分片迁移的开销,并且有效提升分片后的网络安全性;提出了一种基于信誉的改进Raft共识算法(reputation-enhanced Raft,RERaft)用于片内共识,使其拥有一定程度的拜占庭容错能力。实验结果表明,RERaft算法在具备一定拜占庭容错能力的同时可以有效提升系统的共识效率,相比于常见的拜占庭容错算法,具有更好的可扩展性。

基于差异化和空间约束的自动图像描述模型

多头注意力机制是图像描述模型的常用方法,该机制通过多分支结构构建关于输入特征的独特属性,以提高特征模型的区分性。然而,不同分支的独立性导致建模存在冗余性。同时,注意力机制会关注于不重要的图像区域,导致描述的文本不够准确。针对上述问题,提出一种损失函数作为训练目标的正则化项,以提高多头注意力机制的多样性和准确性。在多样性方面,提出一种多头注意力的差异化正则,鼓励多头注意力机制的不同分支关注于所描述目标的不同部件,使不同分支的建模目标变得简单。同时,不同分支相互融合,最后形成完整且更有区分性的视觉描述。在准确性方面,设计一种空间一致性正则。通过建模多头注意力机制的空间关联,鼓励注意力机制关注的图像区域尽可能集中,从而抑制背景区域的影响,提高注意力机制的准确性。提出差异化正则和空间一致性正则共同作用的方法,最终提升自动图像描述模型的准确性。所提方法在MS COCO数据集上对模型进行验证,并与多种代表性工作进行对比。实验结果表明:所提方法显著地提高了图像描述的准确性。

维西-乔后断裂带及其周边地区震群和前震序列活动特征

使用云南地震台网2009—2021年的地震目录和波形数据,采用双差定位方法、基于P波初动极性和S/P振幅比的HASH方法和最大似然法等,获取了维西—乔后断裂带及附近地区M_(L)≥4.0震群和前震序列精确的震源位置、震源机制解、视应力和b值等地震活动参数,对比分析了维西—乔后断裂带及其周边震群及前震序列的活动特征。结果显示:除前震序列频次呈现出反大森定律外,前震序列与普通震群在空间分布、震源机制一致性、视应力和b值等单一参数方面并无明显的差异;但前震序列同时具备震中优势分布方向与区域主要构造展布方向一致、震源机制一致性较好、低b值等特征,建议综合使用时空演化特征、震群参数识别前震序列。

基于一致性算法的区域网络集中式共享储能共享机制

现阶段共享储能市场的交易机制存在准入门槛高、交易成本高、交易过程复杂等问题。在高渗透率新能源接入背景下,传统集中式调度方法存在调度中心通信压力大、可靠性差等问题,而分布式调度方法能通过降低市场交易成本来提高市场效率。针对多座新能源场站与共享储能的标准纳什博弈关系,提出了基于价格一致性的共享储能实时共享机制。基于共享储能的实时交易机制,构建了新能源场站与共享储能运营商的分布式个体决策模型;为了实现去中心化共享,在区域共享网络中提出了基于一致性算法的共享储能实时共享机制,参与者通过交互共享电价达成价格认同,最终实现共享储能社会福利最大化。以新能源场站-共享储能网络为算例进行仿真,验证所提分布式调度方法能达到与集中式调度方法相同的效果,且适用性更优。

融合注意力的教师互一致性半监督医学图像分割

医学图像分割在疾病辅助诊断中起着关键的作用。现有的深度分割模型需要依赖带有标注的数据完成大规模训练,而医学影像标注需要具有专业背景的临床医生进行像素级标注,导致标注数据获取困难。基于半监督的医学图像分割方法利用少量的标注数据和大量的未标注数据进行学习,可以在一定程度上缓解标注数据获取困难的问题。针对半监督分割模型不能充分利用未标注数据中的可学习信息的问题,提出一种半监督分割模型TCA-Net。该模型使用U-Net作为骨干网络,通过在U-Net中引入卷积块注意力模块(CBAM)与多头自注意力模块(MHA)来解决其在下采样过程中的信息丢失问题;为了充分利用未标注数据中的不确定性信息,构建一个教师互一致性模型,该模型由具有1个编码器和3个略有不同的解码器的学生模型与教师模型组成,通过在学生模型的概率映射与教师模型的伪标签之间添加一致性约束,以此在训练过程中最小化输出之间的差异,从而提升模型的分割效果。在公开的WORD腹部多器官数据集与ACDC心脏数据集上进行实验,结果表明,在使用20%标注数据的WORD数据集上,TCA-Net的Dice系数、Jaccard指数、HD95和ASD分别达到90.81%、83.79%、21.38和6.08,在ACDC数据集上分别达到89.69%、81.94%、1.66和0.45。消融实验与对比实验结果表明,TCA-Net能够有效提升未标注数据的利用率,在不同数据集上均达到了较好的分割效果,验证了模型的鲁棒性。