神经网络全局势函数在多相催化中的应用

当今的多相催化研究需要新的技术和方法从原子尺度上表征活性中心结构和反应中间体。本文作者课题组近期开发了理论模拟新技术来探索催化剂活性位点结构，即基于神经网络势函数的大规模原子模拟（LASP）软件中实现的全局神经网络势函数计算方法。本文介绍了该方法可以显著降低催化体系的计算代价，而维持与密度泛函理论同一级别的计算精度，从而解决多相催化中的许多复杂问题。本文对神经网络势函数方法的实现细节和目前已实现的应用场景进行了详细介绍。神经网络势函数可以用来预测材料晶体结构，理解高压氢化条件下TiO₂表面的结构演化和确定三元氧化物ZnCrO晶相中合成气制甲醇活性位点。最后文章分析了神经网络势函数的局限性和今后可能的三个研究方向，即材料性质预测、多元素体系神经网络势函数构造和化学反应拟合。     

基于梯度分布调节策略的Xgboost算法优化

为了解决集成学习模型Xgboost在二分类问题中少数类检出率低的问题，提出了基于梯度分布调节策略的改进的Xgboost算法——LCGHA-Xgboost。首先，通过定义损失贡献（LC）来模拟Xgboost算法中样本个体的损失量；而后，通过定义损失贡献密度（LCD）来衡量Xgboost算法中样本被正确分类的难易程度；最后，提出了梯度分布调节算法LCGHA，依据LCD动态调整样本个体的一阶梯度分布，间接地增大难分样本（主要存在于少数类中）的损失量，减小易分样本（主要存在于多数类中）的损失量，使Xgboost算法偏向对难分样本的学习。实验结果表明,与Xgboost、GBDT、随机森林（Random_Forest）这三大集成学习算法相比，LCGHA-Xgboost算法在多个UCI数据集上的召回率（Recall）值有5.4%~16.7%的提高，AUC值有0.94%~7.41%的提高；在垃圾网页数据集WebSpam-UK2007和DC2010数据集上所提算法的Recall值更是有44.4%~383.3%的提高，AUC值有5.8%~35.6%的提高。LCGHA-Xgboost算法可以有效提高对少数类的分类检出能力，减小少数类的分类错误率。     

Photoactive Earth‐Abundant Iron Pyrite Catalysts for Electrocatalytic Nitrogen Reduction Reaction

The generation of ammonia, hydrogen production, and nitrogen purification are considered as energy intensive processes accompanied with large amounts of CO₂ emission. An electrochemical method assisted by photoenergy is widely utilized for the chemical energy conversion. In this work, earth‐abundant iron pyrite (FeS₂) nanocrystals grown on carbon fiber paper (FeS₂/CFP) are found to be an electrochemical and photoactive catalyst for nitrogen reduction reaction under ambient temperature and pressure. The electrochemical results reveal that FeS₂/CFP achieves a high Faradaic efficiency (FE) of ≈14.14% and NH₃ yield rate of ≈0.096 µg min^?1 at ?0.6 V versus RHE electrode in 0.25 m LiClO₄. During the electrochemical catalytic reaction, the crystal structure of FeS₂/CFP remains in the cubic pyrite phase, as analyzed by in situ X‐ray diffraction measurements. With near‐infrared laser irradiation (808 nm), the NH₃ yield rate of the FeS₂/CFP catalyst can be slightly improved to 0.1 µg min^?1 with high FE of 14.57%. Furthermore, density functional theory calculations demonstrate that the N₂ molecule has strong chemical adsorption energy on the iron atom of FeS₂. Overall, iron pyrite‐based materials have proven to be a potential electrocatalyst with photoactive behavior for ammonia production in practical applications.     

基于模糊谱聚类的不确定蛋白质相互作用网络功能模块挖掘

针对谱聚类融合模糊C-means（FCM）聚类的蛋白质相互作用（PPI）网络功能模块挖掘方法准确率不高、执行效率较低和易受假阳性影响的问题，提出一种基于模糊谱聚类的不确定PPI网络功能模块挖掘（FSC-FM）方法。首先，构建一个不确定PPI网络模型，使用边聚集系数给每一条蛋白质交互作用赋予一个存在概率测度，克服假阳性对实验结果的影响；第二，利用基于边聚集系数流行距离（FEC）策略改进谱聚类中的相似度计算，解决谱聚类算法对尺度参数敏感的问题，进而利用谱聚类算法对不确定PPI网络数据进行预处理，降低数据的维数，提高聚类的准确率；第三，设计基于密度的概率中心选取策略（DPCS）解决模糊C-means算法对初始聚类中心和聚类数目敏感的问题，并对预处理后的PPI数据进行FCM聚类，提高聚类的执行效率以及灵敏度；最后，采用改进的边期望稠密度（EED）对挖掘出的蛋白质功能模块进行过滤。在酵母菌DIP数据集上运行各个算法可知，FSC-FM与基于不确定图模型的检测蛋白质复合物（DCU）算法相比，F-measure值提高了27.92%，执行效率提高了27.92%；与在动态蛋白质相互作用网络中识别复合物的方法（CDUN）、演化算法（EA）、医学基因或蛋白质预测算法（MGPPA）相比也有更高的F-measure值和执行效率。实验结果表明，在不确定PPI网络中，FSC-FM适合用于功能模块的挖掘。     

Frequency dependence of plasma characteristics at different pressures in cylindrical inductively coupled plasma source

The effects of driving frequency on plasma parameters and electron heating efficiency are studied in cylindrical inductively coupled plasma (ICP) source. Measurements are made in an Ar discharge for driving frequency at 13.56/2 MHz, and pressures of 0.4–1.2 Pa. In 13.56 MHz discharge, higher electron density (n_e) and higher electron temperature (T_e) are observed in comparison with 2 MHz discharge at 0.6–1.2 Pa. However, slightly highern_e andT_e are observed in 2 MHz discharge at 0.4 Pa. This observation is explained by enhanced electron heating efficiency due to the resonance between the oscillation of 2 MHz electromagnetic field and electron-neutral collision process at 0.4 Pa. It is also found that the variation ofT_e distribution is different in 13.56 and 2 MHz discharge. For ICP at 13.56 MHz, T_e shows an edge-high profile at 0.4–1.2 Pa. For 2 MHz discharge,T_e remains an edge-high distribution at 0.4–0.8 Pa. However, the distribution pattern involves into a center-high profile at 0.9–1.2 Pa. The spatial profiles ofn_e remain a center-high shape in both 13.56 and 2 MHz discharges, which indicates the nonlocal kinetics at low pressures. Better uniformity could be achieved by using 2 MHz discharge. The effects of gas pressure on plasma parameters are also examined. An increase in gas pressure necessitates the rise ofn_e in both 13.56 and 2 MHz discharges. Meanwhile, T_e drops when gas pressure increases and shows a flatter distribution at higher pressure.     

The reliability of asset management regime of the SROH using air void content of asphalt mixtures

The comparison pairs of cores (each 100 mm apart) from 68 reinstatements sites from various parts of the UK revealed that, the compounding consequences of generic non-homogeneous characteristics of hand laid recipe mixed materials (specified in Specification for the Reinstatement of Openings in Highways (SROH)) and high likeliness of being biased during air void (AV) testing makes the coring method extremely unreliable with very low repeatability and reproducibility. The wide-ranging maximum density reported in every instance in the comparison pair coring experiments meaningfully rationalizes the distorted homogeneity of materials. Although not only maximum density but also bulk density of adjacent cores located only 100 mm apart were found to be varied in the case of every pair in this study. Furthermore, the in situ performance shown by from 50 reinstatements after experiencing 1.5–10 years real-life ageing from various parts of the UK predictably indicates that either the linkage between the reinstatement with non-compliant AV and its impact on footways durability is non-proven or the reported AV content is extremely over estimated. At 95% level of significance, there exists enough evidence to conclude that, due to high uncertainty, very low repeatability and reproducibility and poor reliability with high chances of bias, the assessment of hand laid reinstatement work by AV testing will expose both the contractor and the client to unacceptable risk.     

Stabilizing temperature-capacitance dependence of (Sr,Pb, Bi)TiO3-Bi4Ti3O12 solutions for energy storage

(1-x)Sr_0.7Pb_0.15Bi_0.1TiO₃-xBi₄Ti₃O₁₂ ((1-x)SPBT-xBIT, x = 0-0.125) bulk ceramics were developed and calcined via the solid-state method, aimed at the application of pulsed power capacitors. The phase structures, temperature stability, hysteresis loop, and discharge properties were systematically investigated. Considering both the temperature stability and dielectric properties, 0.925SPBT-0.075BIT bulk ceramics with a capacitance variation satisfying the X7R specification were developed for pulsed power capacitors. The energy storage density was 0.252 J/cm³, and the ceramics showed high temperature stability at 80 kV/cm. The discharge current waveforms of the 0.925SPBT-0.075BIT ceramics were recorded. A high discharge power density of approximately 1.01 × 10⁸ W/kg with an 8 Ω load resistor and short discharge period of 84 ns were achieved at 50 kV/cm. The good temperature stability properties and high power density show that the 0.925SPBT-0.075BIT ceramics are well suited for pulsed power capacitors with a wide temperature range.     

Microstructural and Tribological Behavior of Pulse Electroplated Nickel Barrier on Copper Conductors

The microstructural, mechanical, and tribological behaviors of electroplated Ni on Cu conducting substrates have been investigated in this study. The microstructural studies were performed by X-ray diffraction (XRD) analysis and scanning electron microscopy (SEM). The results showed that initially (111) with (220), (200) Ni texture components were predominant in the coating, and increasing the current density from 0.1 to 0.5 A/cm² led to the development of a strong (111) texture. The presence of ultrafine grains coupled with a (111) Ni texture improved the coating microhardness and wear properties significantly. It was shown that with an increase in current density, wear resistance of the coatings improved significantly and the electrical resistivity increased due to the highly populated grain boundaries.     

Contribution of metal vapor mass at periphery part of pulsed arc to electromagnetic force in weld pool

The arc welding has been used in various welding methods because it is inexpensive and high in strength after welding. However, it is a problem that accidents such as collapse of the bridge occur because of the welding defects. The welding of low cost and high productivity is required without the welding defects. The pulsed TIG welding is inexpensive and capable of high‐quality welding. The electromagnetic force contributing to penetration changes because the transient response of arc temperature and iron vapor generated from anode occurs. However, the analysis of pulsed TIG welding with metal vapor has been elucidated only metal vapor concentration near anode with transient phenomenon and heat flux. Thus, the theoretical elucidation of penetration depth with control factor has not been researched. In this paper, the contribution of metal vapor mass at the periphery part of pulsed arc to the electromagnetic force in the weld pool is elucidated. As a result, the iron vapor mass at periphery part decreased with increasing the frequency. The iron vapor was stagnated at axial center within one cycle. The electromagnetic force to the penetration depth direction in weld pool increased at axial center. Therefore, the metal vapor mass at periphery part plays an important role for the electromagnetic force increment at axial center.     

一种改进的最小二乘回归子空间分割方法

最小二乘回归(LSR)算法是一种常见的子空间分割方法，由于LSR具有解析解，因此它的聚类性能较高。然而LSR算法是应用谱聚类方法聚类数据，谱聚类方法初始化聚类中心是随机的，会影响后面的聚类效果。针对这一问题，提出一种基于聚类中心局部密度和距离这2个特点的改进的LSR算法（LSR-DC）。在Extended Yale B数据集上进行实验，结果表明，该算法有较高的聚类精度，具有一定的鲁棒性，优于现有LSR等子空间分割方法。