基于语言模型的预训练技术研究综述

预训练技术当前在自然语言处理领域占有举足轻重的位置。尤其近两年提出的ELMo、GTP、BERT、XLNet、T5、GTP-3等预训练模型的成功,进一步将预训练技术推向了研究高潮。该文从语言模型、特征抽取器、上下文表征、词表征四个方面对现存的主要预训练技术进行了分析和分类,并分析了当前自然语言处理中的预训练技术面临的主要问题和发展趋势。     

Vacancies and phosphorus atoms assembled in amorphous urchin-like Co3O4 for highly efficient overall water splitting

Designing highly efficient and low-cost electrocatalysts is essential for water splitting. Herein, urchin-like Co₃O₄ microspheres are firstly grown on nickel foam by a hydrothermal method, then Oxygen vacancies, phosphorus doping are effectively assembled in Co₃O₄ electrocatalysts. The introduction of oxygen vacancies and phosphorus doping will adjust the electronic structure of Co which increase the intrinsic catalytic activity and improve the adsorption energy of intermediates, simultaneously, progressively transform the crystal into randomly arranged atoms structure with short range order resulting in more active sites participate in the catalytic reaction. Moreover, the catalyst of vacancies Co₃O₄-Ov and phosphorus doping Co₃O₄–P demonstrate excellent performance in oxygen evolution reaction (OER) and hydrogen evolution reaction (HER) in alkaline media, Co₃O₄-Ov sample served as anode while Co₃O₄–P as cathode to form an electrolytic cell needs only 1.58 V to reach 20 mA cm^?2 for overall water splitting.     

Human responses to high levels of carbon dioxide and air temperature

In this study, 30 subjects were exposed to different combinations of air temperature (T_a: 24, 27, and 30°C) and CO₂ level (8000, 10 000, and 12 000 ppm) in a high-humidity (RH: 85%) underground climate chamber. Subjective assessments, physiological responses, and cognitive performance were investigated. The results showed that as compared with exposure to T_a = 24°C, exposure to 30°C at all CO₂ levels caused subjects to feel uncomfortably warm and experience stronger odor intensity, while increased mental effort and greater intensity of acute health symptoms were reported. However, no significant effects of T_a on task performance or physiological responses were found. This indicated that subjects had to exert more effort to maintain their performance in an uncomfortably warm environment. Increasing CO₂ from 8000 to 12 000 ppm at all T_a caused subjects to report higher rates of headache, fatigue, agitation, and feeling depressed, although the results were statistically significant only at 24 and 27°C. The text typing performance and systolic blood pressure (SBP) decreased significantly at this exposure, whereas diastolic blood pressure (DBP) and thermal discomfort increased significantly. These effects suggest higher arousal/stress. No significant interaction effect of T_a and CO₂ concentration on human responses was identified.     

牵引逆变器IGBT风冷散热试验及仿真平台搭建

IGBT散热性能是影响高速动车组在行驶过程中安全性、可靠性及动力性的重要因素。本文以某出口铰接车牵引变流器箱为试验研究载体,研制了专用风冷散热试验台;同时运用仿真软件Flotherm进行了仿真计算,通过实物试验与仿真计算的对比研究,验证了仿真计算的可行性。     

主回路滞流分支管振动超标的原因分析与设计优化

在某核电厂热态功能试验中，巡视发现主回路的某滞流分支管振动超标。根据实测振动随温度变化的趋势、管道固有声模态分析和模态分析等，推测管道内流体的声振动激励引发管道共振。根据滞流管三通处流体的漩涡脱落频率计算和声振动频率计算，推测漩涡脱落与滞流管内流体的固有声振动频率锁定，引发声振动激励放大。综合分析计算和实测结果，确定滞流管振动超标的原因为流声固耦合振动。对管道支架进行了优化，以错开声固耦合共振频率，在三通处倒圆角削弱漩涡脱落，以降低流声耦合的声振动幅值。优化后的分支管振动会明显降低，以保证承压边界的安全。     

高能同步辐射光源逐束团束流位置测量电子学研制

基于高能同步辐射光源（HEPS）储存环，研制了一套逐束团束流位置测量（BPM）电子学系统，电子学的硬件部分由模拟信号采集板卡和数字信号处理板卡组成，软件部分由底层固件和顶层应用软件组成。系统的采样频率为500 MHz，带宽为1 GHz，对来自储存环BPM探头的4路模拟信号进行数字化，得到束团幅度数据，利用ZYNQ芯片计算出每个束团在真空管道中的位置。逐束团BPM电子学在实验室的测试结果为：输入信号峰峰值小于1.8 V时ADC通道非线性度小于1%，无杂散动态范围约60 dB，灵敏度系数取8.26 mm时位置分辨率优于10 μm，测试结果满足HEPS逐束团BPM的需求。     

Facile in-situ Solvothermal Method to synthesize double shell ZnIn2S4 nanosheets/TiO2 hollow nanosphere with enhanced photocatalytic activities

The design and construction of efficient visible light responsive composite photocatalysts with intimate interfacial contacts in photocatalytic field have attracted huge interest. Herein, a double-shelled ZnIn₂S₄ nanosheets/TiO₂ hollow composite single nanosphere (ZIS/TiO₂) was first fabricated by a facile hydrothermal process, where 2D ZnIn₂S₄ nanosheets self-assembled on the external surface of TiO₂ hollow nanosphere to form the double-shelled hollow single sphere. The morphologies, structures, optical properties of as-prepared double-shelled ZIS/TiO₂ hollow nanospheres were characterized in detail. The photocatalytic activities of double-shelled ZIS/TiO₂ nanospheres for the photodegradations of Tetracycline hydrochloride, Levofloxacin and Rhodamine B under visible light irradiation have been investigated. Compared to pure TiO₂ and ZnIn₂S_4, the obtained ZIS/TiO₂ samples have significantly improved photocatalytic performances. The most optimal photocatalytic activity of ZIS/TiO₂-2 nanocomposite with 64 wt% ZnIn₂S₄ nanosheets coated is observed, and its degradation rate constant is 2.32 and 2.14 times as high as those of pure TiO₂ and ZnIn₂S₄. The superior photocatalytic performance of ZIS/TiO₂ nanocomposite can be ascribed to its unique double shell hollow structure and the synergistic effect between ZnIn₂S₄ and TiO₂. Our result provides some guidance for designing novel morphologies of composite photocatalyst with good photocatalytic performance.     

基于LCOE最优的数字化风电场及其关键技术研究

数字化风电场是风电场综合利用数字化技术的新阶段。在分析风电场建设管理所面临的挑战的基础上，讨论了平准化度电成本（LCOE）的计算公式和内涵，给出了数字化风电场的定义，阐述了基于LCOE最优的数字化风电场的“风、机、场、网、环”五大特征以及数字化技术在降低LCOE中的作用。最后，总结了实现数字化风电场的风资源、大数据平台、数据分析建模、数字孪生和数据安全五大关键技术。     

基于ＰＩＤ控制算法的直流微网容错控制方法研究

构建直流微网容错控制对象模型，调节直流微电网的输出回路参数；以输出功率、直流微网的 参考电压、弱电网下系统惯性响应特征等为约束参量，构建直流微网容错控制目标函数，在不同电网强度下 进行直流微网容错控制的参数自整定性调节，采用无功环比例积分控制方法进行直流微网容错寻优分析， 建立模糊 ＰＩＤ控制模型，采用变结构的模糊 ＰＩＤ控制方法进行直流微网容错控制过程中的自适应加权学习 和误差反馈调节，实现直流微网容错控制改进设计。仿真结果表明，采用该方法进行直流微网控制的容错 性能较好，输出稳定性较强，具有较好的直流微网输出增益。