基于图神经网络的网络性能智能预测

传统网络性能预测技术存在网络状态获取不够全面及网络性能评估准确性欠佳等问题，利用图神经网络学习推理网络关系数据的特点，结合捕获的网络全局信息，提出了一种基于图神经网络的网络性能智能预测方法。通过网络系统抽象及网络性能建模，将复杂的网络信息转化为形式化的图数据进行描述，利用图空域卷积处理图网络节点的消息传递过程，实现网络信息之间的关系推理，研究了实现网络性能预测的图神经网络模型，提出了一种能处理流量矩阵、网络拓扑、路由策略、节点配置的图神经网络体系结构，最后通过实验论证了模型能更好地实现包括时延、抖动和丢包率的网络性能的准确预测。     

基于特征建模的TDD系统大气波导干扰控制体系研究

大气波导干扰是特定气象条件下发生的时分双工（time-division duplex，TDD）系统内干扰，是TDD移动通信系统大规模组网面临的顽疾。在总结分析大气波导干扰成因和分类等的基础上，对大气波导干扰进行建模和表征，验证了海量干扰源在时域和频域的功率集总特征，并结合大量4G/5G现网实测数据给出了典型条件下内陆波导和海面波导的量化干扰信号传播模型，对于干扰的预测和预防具有重要意义。基于干扰特征，给出了TDD系统预防大气波导干扰的帧结构与组网的4项设计原则，5G现网数据表明干扰控制方案有效，上行干扰下降10 dB以上，相关原则对于6G系统的设计也具有指导意义。     

Predetermination of potential plastic hinges on reinforced concrete frames using GFRP reinforcement

In the past, glass fiber-reinforced polymer (GFRP)-reinforcement has been successfully applied in reinforced concrete (RC) structures where corrosion resistance, electromagnetic neutrality, or cuttability were required. Previous investigations suggest that the application of GFRP in RC structures could be advantageous in areas with seismic activity due to their high deformability and strength. However, especially the low modulus of elasticity of GFRP limited its wide application as GFRP-reinforced members usually exhibit considerably larger deformations under service loads than comparable steel-reinforced elements. To overcome the aforementioned issues, the combination of steel and GFRP reinforcement in hybrid RC sections has been investigated in the past. Based on this idea, this paper presents a novel concept for the predetermination of potential plastic hinges in RC frames using GFRP reinforcement. To analyze the efficiency of the concept, nonlinear finite element simulations were performed. The results underscore the high efficiency of hybrid steel-GFRP RC sections for predetermining potential plastic hinges on RC frames. The results also indicate that the overall seismic behavior of RC structures could be improved by means of GFRP as both the column base shear force during the seismic activity as well as the plastic deformations after the earthquake were considerably less pronounced than in the steel-reinforced reference structure.     

基于SRAM的通用存算一体架构平台在物联网中的应用

最近,存算一体(IMC)架构引起了广泛关注,并被认为有望成为突破冯诺依曼瓶颈的新型计算机架构,特别是在数据密集型(data-intensive)计算中能够带来显著的性能和功耗优势.其中,基于SRAM的IMC架构方案也被大量研究与应用.该文在一款基于SRAM的通用存算一体架构平台——DM-IMCA的基础上,探索IMC架构在物联网领域中的应用价值.具体来说,该文选取了物联网中包括信息安全、二值神经网络和图像处理在内的多个轻量级数据密集型应用,对算法进行分析或拆分,并将关键算法映射到DM-IMCA中的SRAM中,以达到加速应用计算的目的.实验结果显示,与基于传统冯诺依曼架构的基准系统相比,利用DM-IMCA来实现物联网中的轻量级计算密集型应用,可获得高达24倍的计算加速比.     

基于形态特征提取的急性下壁心肌梗死BiLSTM网络辅助诊断算法

急性下壁心肌梗死是一种病发急、进展快、致死率高的心脏疾病,该文提出一种新颖的基于形态特征提取的BiLSTM神经网络分类的急性下壁心肌梗死辅助诊断算法,可大幅度提高医生对急性下壁心肌梗死疾病的诊断效率并有助于及时确诊.算法包括:对胸痛中心数据库心拍信号进行降噪及心拍分割;根据临床心内科医学诊断指南提取了12导联波形距离特征和分导联波形幅值特征;依据提取的特征搭建LSTM与BiLSTM神经网络进行心拍的分类识别;使用PTB公开数据库和胸痛中心数据库多临床中心进行交叉验证.实验结果表明,加入胸痛中心真实临床数据后,基于形态特征提取BiLSTM神经网络的急性下壁心肌梗死辅助诊断算法准确率达到99.72％,精度达到99.53％,灵敏度达到100.00％,同时F1-Score达到99.76.该算法比其他现有算法准确率提高至少1％,该项研究具有非常重要的临床应用价值.