混凝土裂缝是影响混凝土结构服役性能的重要因素之一，研究开发一种引入生成对抗网络和注意力机制的SSD目标检测修正算法，对精准识别裂缝目标具有重要意义。通过建立全卷积神经网络与生成对抗网络(GAN)相结合的改进网络结构，以工程采集的有限裂缝样本图像为基础生成高质量图像数据以解决数据集获取难的问题；同时引入混合注意力机制(CBAM)以增强SSD特征提取网络对较小裂缝的感知能力。研究结果表明：使用引入注意力和生成对抗网络的目标检测算法检测混凝土表面裂缝，在裂缝识别时各项性能指标均超过了83%,平均准确率AP更是达到了91.51%,相较于原始的SSD目标检测算法提高了10.36%。

盐渍脆梅因其丰富的营养物质及爽脆的口感而广受消费者喜爱，然而由于加工过程中杀菌不彻底，经常会在货架期内因染菌而引起胀袋现象，造成食品安全问题和经济损失，因此研究可高效抑菌并适用于工业化生产的杀菌方法意义重大。对胀袋盐渍脆梅中的污染微生物进行分离、鉴定，并研究了防腐剂与非热杀菌技术对其抑制效果，确定能有效抑制污染菌生长的技术方法。研究结果表明：污染微生物为拜氏接合酵母，0.78 mg/mL的山梨酸钾和苯甲酸钠均不能抑制拜氏接合酵母，ε-聚赖氨酸盐酸盐和纳他霉素对其最小抑制质量浓度均为0.10 mg/mL。低温等离子体、脉冲强光和酸性氧化电位水处理等3种非热杀菌技术不仅能有效抑制酵母菌，而且不会对盐渍脆梅产品产生负面的感官影响。研究结果可为盐渍脆梅的实际生产提供理论依据和技术支持。

偏氟乙烯和三氟乙烯的共聚物（Ｐ（ＶＤＦ－ＴｒＦＥ））由于具有低密度、柔性、低阻抗以及生物相容 性好等优点，在生物医学、电子器件、能源存储等领域具有广泛应用前景。然而相较于陶瓷压电材 料，Ｐ（ＶＤＦ－ＴｒＦＥ）薄膜的极化值小，压电常数也较低。为提高Ｐ（ＶＤＦ－ＴｒＦＥ）的压电性能，需要采 用有效途径来提高压电相β相的质量分数以及薄膜结晶度。采用高温固相法在１　３５０℃下烧结获 得处于ＭＰＢ结构的０．４Ｂａ（Ｚｒ０．２Ｔｉ０．８ ）Ｏ３－０．６（Ｂａ０．７Ｃａ０．３ ）ＴｉＯ３ （ＢＣＺＴ）陶瓷颗粒，用盐酸多巴胺 （ＰＤＡ）对ＢＣＺＴ进行表面处理后以其为填料与Ｐ（ＶＤＦ－ＴｒＦＥ）共混形成复合薄膜。实验结果表 明，经表面处理的ＰＤＡ＠ＢＣＺＴ与Ｐ（ＶＤＦ－ＴｒＦＥ）基质的界面相容性得到了明显改善，复合薄膜的 结晶度和压电性能大幅提高。当加入质量分数（湿重）为５％的ＰＤＡ＠ＢＣＺＴ时，复合薄膜的结晶 度由未添加时的４７．６０％提高到５９．４９％，压电系数（ｄ３３ ）由１０．２ｐＣ／Ｎ提高到１９．９ｐＣ／Ｎ。将添 加质量分数（湿重）为５％的ＰＤＡ＠ＢＣＺＴ复合薄膜制成压电器件，用约５Ｎ的力敲击薄膜器件可 以产生２０Ｖ左右的电压并成功点亮发光二极管。

基于中药质量标志物（Ｑｕａｌｉｔｙ　ｍａｒｋｅｒ，Ｑ－ｍａｒｋｅｒ）的“五原则”，综合辨析抗衰老口服液中的 关键质量标志物。采用ＨＰＬＣ－Ｑ－ＴＯＦ－ＭＳ／ＭＳ技术共鉴定出抗衰老口服液中１０３个化合物。结 合网络药理学筛选与“抗衰老”药效相关的活性成分，预测获得药物靶点４５２个，疾病靶点７６９个， 共有靶点９６个以及基因通路１５８条。根据Ｑ－ｍａｒｋｅｒ的有效性、配伍性、特有性、可测性和溯源性 “五原则”，最终优选出毛蕊花糖苷、Ｒｈ４ 、梓醇、地黄苷Ｄ、Ｒｈ１ 、２０－（Ｓ）－Ｒｇ３ 、Ｒｆ、Ｒｇ２ 、焦地黄苯乙醇 苷Ａ１ 和洋地黄叶苷Ｃ作为抗衰老口服液的关键Ｑ－ｍａｒｋｅｒｓ，为抗衰老口服液的整体质量控制和临 床应用提供一定参考。

压实黏土具有较好的防渗性能，常被用作卫生垃圾填埋场的衬垫材料，然而其抗裂性能较差， 且在衬垫变形开裂后其防渗性能也进一步降低。笔者设计了一款模拟衬垫变形和渗透的试验装 置，研究衬垫变形条件下其渗透性能的变化规律，并采用粒子图像测速技术计算分析衬垫表面土体 的位移规律。试验结果表明：当变形量为１，２，３ｃｍ时，衬垫产生裂隙，裂隙宽度分别为１，７， １５ｍｍ，衬垫破坏水头分别为４．６，４．２，３．６ｍ；衬垫变形时衬垫厚度对渗透性能有较大影响，当变 形量为３ｃｍ时，厚度２０ｃｍ的衬垫被裂隙贯穿，厚度３０，４０ｃｍ的衬垫上部无裂隙区域厚度分别为 ８，２０ｃｍ。

For the distributed photovoltaic (PV) grid-connected security assessment, the assessment index system including equipment and system security is designed, and a comprehensive assessment method based on the hierarchical analysis process (AHP) and the technique for order preference by similarity to the ideal solution (TOPSIS) is proposed. Firstly, taking into account the characteristics of grid security assessment and the applicability of indicators, the assessment indicator system is constructed from the target layer, criterion layer and indicator layer; secondly, the weights of each indicator are determined by the AHP method; thirdly, the comprehensive evaluation index is calculated and sorted by TOPSIS; and finally, the comprehensive assessment method of security of distributed PV access to the distribution network is determined. The proposed method is validated by taking the actual distribution network of a county in east China as an example. The experimental results show that distributed PV access to the distribution network optimises the performance of low voltage risk and comprehensive line loss rate as well as voltage deviation. While the power reversal risk, overvoltage risk, and effective coverage of PV inverter protection devices decreased relative to the pre-access performance. In a comprehensive comparison, distributed PV grid connection improves distribution network security.

How to effectively analyze stock-related text data and provide stock recommendations to investors is one of the important research directions in the field of intelligent investment advisors. Most of the existing methods that rely on text analysis such as research reports to make stock recommendations only focus on the impact of the text as a whole on the stock and neglects the fact that there may be multiple events in the text. Multiple events interact with each other and may have a more complex impact on the stock, which cannot be taken into account by using only the text as a whole. To address this problem, a stock recommendation method based on event extraction is proposed. Not only the different events appeared in the text are taken into account, but also the temporal relationship between the events are considered. First, the stock text data is obtained and the events are extracted through the cloud-based large model. Second, the temporal relationships between events are determined pairwise through the local large model and form an event-time triplet. Finally, the stock text and the event-time triplet collection are spliced together to classify and predict the rise and fall of stocks. The hierarchical backtesting is performed based on the factors constructed by the classification probability. The experimental results prove that the method can not only extract the event information in the stock text, but also determine the temporal relationship between events. At the same time, the method can improve the gain of hierarchical backtesting, which contributes to more accurate recommendation of stocks. Compared with the traditional approach, the total simulated gain is improved by 0.0323.

The lower limb rehabilitation exoskeleton robot system has been widely used in the fields of healthcare and disability rehabilitation due to its ability to provide intelligent rehabilitation methods for patients. Faced with the rapid development of the current robotics industry and shortcomings in classical control strategies, controllers combining various intelligent strategies have become the mainstream solution. However, intelligent strategies such as neural networks, model predictive control, and iterative control are difficult to put into practical application under the current hardware conditions of the system. Compared with the above methods, sliding mode control can not only achieve better control accuracy, but also has higher application feasibility. Based on this, an adaptive recursive terminal sliding mode control method is designed for trajectory tracking control of lower limb rehabilitation exoskeleton robots. Firstly, a mechanism dynamics model is established for the lower limb rehabilitation exoskeleton robot based on the Lagrangian modeling method. Then, a recursive terminal sliding mode surface is constructed for the dynamic model, and an adaptive sliding mode controller is designed. The stability of the proposed algorithm is proved in the sense of Lyapunov. Finally, the effectiveness and superiority of the proposed method are verified through comparative experiments with non-singular terminal sliding mode on the lower limb rehabilitation exoskeleton robot experimental platform.

碳纳米管(CNTs)是一种具有优异力学与电学性能的新型碳材料，适量的碳纳米管即可使几乎绝缘的水泥基材料形成具有良好压阻性能的传感器。而压阻性能决定了传感器灵敏度大小，其受到多类因素的影响。针对这一问题，综述了国内外对碳纳米管水泥基传感器材料组构、受载特性和微观模型的研究，从多方面阐述并总结了目前碳纳米管水泥基传感器压阻性能的研究进展，针对部分研究的不足提出问题，分析原因并进行了讨论，以期为碳纳米管水泥基传感器的设计与应用提供参考。

有效分组是提升协作学习效率的关键，合理的分组能够使个体和组内成员均获得最大化的习得成效；然而，目前的分组方法未对学习者知识水平特征深度计算，无法保证组内成员在知识结构上相辅相成。因此，提出一种融入深度知识追踪优化模型(DKVMN-KT)的协作学习分组方法。首先，采用深度知识追踪优化模型对学习者的知识状态建模，得到学习者的知识掌握程度；然后，利用K-means方法对所有学习者进行相似聚类，最后根据分组的异质性原则，将不同簇的学习者分配到适合的学习小组。实验结果表明：该方法能够实现学习者知识结构层面的有效分组，分组结果在知识结构上更具有公平性和异质性。

Polyhalogenated carbazoles (PHCZs) are an emerging class of organic pollutants characterized by persistence, bioaccumulation and biomagnification. These compounds have been widely detected in various environmental media, including soil, sediment, water and air. However, the sources of PHCZs remain unclear. The printing and dyeing industry is suspected to be one of the significant contributors to PHCZ emissions. In this study, a typical printing and dyeing industrial park in China was selected as the research site to investigate the pollution levels of PHCZs and related compounds, including carbazole and benzocarbazole, in environmental samples. The results showed that the target compounds were present at relatively high concentrations, suggesting that their environmental residues may be associated with the activities of the printing and dyeing industry in this area. Although the toxic equivalent quantity (TEQ) of the target compounds was below the safety threshold, potential risks related to bioaccumulation and long-term exposure still warrant attention.

针对电机生产检测环节中对异常检测方法提出的高准确率及高泛化性需求，提出了一种基于 嵌套式深度学习模型的方法来实现电机的异常检测。首先通过卷积模型（Ｃｏｎｖｏｌｕｔｉｏｎａｌ　ｎｅｕｒａｌ ｎｅｔｗｏｒｋｓ，ＣＮＮ）和自注意力模型（Ｓｅｌｆ－ａｔｔｅｎｔｉｏｎ）组成集成学习，提取相邻时序的特征；然后引入残 差网络丰富原始振动特征，以提高模型泛化性；最后通过时序卷积模型（Ｔｅｍｐｏｒａｌ　ｃｏｎｖｏｌｕｔｉｏｎａｌ ｎｅｔｗｏｒｋｓ，ＴＣＮ）输出最终检测结果。实验结果表明：笔者所提出方法的平均准确率达到了９５．９３％， 且模型具有较好的泛化能力，所提出的异常检测方法能很好地在小波变换、傅里叶变换、小波包分 解等特征提取方法无法生效的场景中发挥作用，为工业电机自动化生产提供参考。

为揭示硫铝酸盐水泥（ＳＡＣ）对脱硫建筑石膏－普通硅酸盐水泥－硫铝酸盐水泥组成的三元 胶凝材料自流平砂浆性能的影响规律，研究了ＳＡＣ掺量对自流平砂浆凝结时间、力学性能、干燥收 缩、耐水性和微观结构的影响。试验结果表明：随着ＳＡＣ掺量的增加，砂浆水化放热峰值升高，放 热速率增大，砂浆凝结时间缩短；砂浆的１ｄ强度随ＳＡＣ掺量的增加而提高；当ＳＡＣ掺量大于５％ 时，２８ｄ抗折强度逐渐下降；２８ｄ抗压强度随ＳＡＣ掺量的增加先提高后下降，当ＳＡＣ掺量＞１５％时， 砂浆２８ｄ抗压强度较基准组提升５０％以上；体积膨胀随ＳＡＣ掺量的增加先升高后降低，当ＳＡＣ掺量 为１０％时，膨胀值最大；经时吸水率随ＳＡＣ掺量的增加先降低后略升高，当ＳＡＣ掺量为２５％时，砂浆 ２８ｄ吸水率最小；２８ｄ砂浆试件浸水１ｄ后的强度随ＳＡＣ掺量的增加而提高，随着浸水时间的延长， 含ＳＡＣ砂浆试件的强度持续发展；掺入ＳＡＣ后，水化产物中出现更多的钙矾石与氢氧化铝凝胶。

以油莎豆为原料，对采用Osborne法提取得到的4种蛋白进行了理化特性与降血糖活性研究。实验结果表明：油莎豆蛋白中谷蛋白、清蛋白、球蛋白和醇溶蛋白分别占39.75%,26.98%,8.10%和4.57%。其中谷蛋白具有较好的物理稳定性，持水性为1.82 g/g,持油性为15.33 g/g,乳化性为8.739 m2/g,起泡性为12.5%。体外降血糖活性实验表明：谷蛋白对α-葡萄糖苷酶的抑制活性最好，IC50为2.79 mg/mL。斑马鱼降血糖模型也表明谷蛋白具有较强的体内降血糖活性。研究结果表明油莎豆谷蛋白具有开发成为降血糖功能食品的潜力。

针对设置拉结件的钢板在弹性基底上的局部稳定问题，首先，根据能量变分原理建立了在轴向压力作用下，考虑弹性基底和拉结件拉力的钢板局部屈曲非线性特征方程及其伽辽金表达式；其次，通过Matlab进行牛顿迭代得到了钢板的屈曲应力和变形模态；再次，经过对所得变形曲面的分析，得出了钢板在不同拉结件布置方法下的3种典型屈曲模态；最后，引入板件有效宽度法则，通过瑞利-里兹法推导出了不同模态下组合板的局部失稳临界荷载计算公式，进而得到了组合板在轴向受压状态下纵、横两个方向拉结件最大间距的限制要求。结果表明：当拉结件设置达到4排后非加载边的边界条件对钢板屈曲荷载的影响可以忽略；在拉结件排列密度相同的情况下，当拉结件纵、横向间距之比β为0.6～1.2时，交错排列下的钢板屈曲应力达到最大值，此时板件实际的屈曲系数约为5.8,拉结件正交排列时板件的屈曲系数最大值大致相同，此时其纵、横向间距之比β约为0.6。

Micro/nanoplastics (MPs/NPs) can be absorbed and accumulated by plants, but their impact on plant toxicity, oxidative stress, and nutritional quality remains unclear. This study investigated the effects of polystyrene (PS) particles and polystyrene particles containing phthalic acid esters (PAEs) (PS-PAEs particles) on cell viability, oxidative stress, and nutritional quality in lettuce cells. The results showed that exposure to PS and PS-PAEs particles reduced cell viability by 6.50% and 21.22%, respectively. The mass fractions of hydrogen peroxide and malondialdehyde in the cells increased by 68.60% and 66.74%, and 34.29% and 67.07%, respectively. Meanwhile, the activities of catalase, peroxidase, and superoxide dismutase decreased by 9.60% and 12.52%, 23.36% and 31.37%, and 31.67% and 30.17%, respectively, indicating that the particles caused oxidative damage and inhibited enzyme activities in lettuce cells. Consequently, the mass fractions of vitamin C, soluble sugar, and soluble protein in the cells decreased by 6.88% and 29.37%, 14.51% and 22.60%, and 33.39% and 47.74%, respectively. Compared to PS particles, PS-PAEs particles exhibited greater toxicity to lettuce cells, likely due to the additional damage caused by PAEs released from the PS-PAEs particles. The study suggests that MPs/NPs pose potential risks to plants by inducing oxidative stress, impairing cell viability, and reducing nutritional quality.

曳引钢带电梯具有高舒适、低功耗和节能环保的优势，其应用场景越来越广泛。钢带在断丝、断裂等极限条件下的工作状态对其安全运行有着至关重要的影响，然而现有研究缺少合理有效的曳引钢带分析模型。针对曳引钢带电梯精细化性能分析以及细化数值模拟仿真模型建立的需求，以破断强度为43 kN的曳引钢带为研究对象，对其在受载拉伸状态下进行受力分析，建立曳引钢带的物理模型，通过微机控制电子万能试验机对钢带内钢丝绳进行拉伸试验，测量出数值模拟中所需的材料属性，以此作为曳引钢带在ANSYS数值模拟中的属性参数设定依据，得到曳引钢带数值模拟模型。基于此，对钢丝绳进行细化建模并进行数值模拟试验，得到了仿真条件下的钢带包覆层与内部钢丝绳的应力分布。将数值模拟试验结果与实际拉伸试验结果进行比较，结果表明仿真结果与实测结果相对误差为0.84%,验证了曳引钢带模型的准确性。

为研究X形矩形钢管相贯节点平面外受弯滞回性能，开展了1个节点试件的平面外往复弯矩加载试验。在规范已有节点平面外受弯承载力计算式的基础上，提出考虑主管翼缘和支主管斜交影响的节点承载力修正计算式。研究结果表明：节点试件破坏模式为主管表面塑性软化后撕裂、主管翼缘塑性软化并伴随着支管根部焊缝撕裂，滞回曲线形状饱满；基于简单屈服线模型的已有规范计算式低估了节点受弯承载力，比试验值低34%,比有限元值大多低10%～35%;修正后的节点受弯承载力计算式与试验值较为接近，仅比试验值低3.4%,与有限元值差异大多低于10%;当支主管夹角60°≤θ≤90°时，1/sinθ较为精确地反映了支主管斜交对节点承载力的影响；当30°≤θ≤45°时，1/sinθ则明显低估了支主管斜交对节点承载力的影响。

采用有机溶剂洗涤预处理+热解法处理化工废盐并进行纯化研究，考察了该方法对有机物的脱除效果。研究结果表明：有机溶剂种类和洗涤方式对于废盐热解脱除有机物效果影响很大，采用甲苯与废盐质量比为1.0∶1.5振荡洗涤20 min后，在升温速率15℃/min、热解温度600℃、热解2.5 h条件下，工业废盐中的有机物基本完全去除，再经水洗、过滤及重结晶后可得到高纯度副产物工业盐，实现化工废盐的资源化应用。

采用斜风下大跨度桥梁三维全模态颤振分析程序，分析1 400 m主跨的三塔斜拉-悬吊协作体桥设计方案的动力特性和斜风下的颤振稳定性，并与相同主跨的双塔斜拉-悬吊协作体系桥和三塔悬索桥进行对比分析，揭示其颤振性能及主要影响因素。研究结果表明：三塔斜拉-悬吊协作体系桥结构柔性大，竖向刚度和横向刚度均很小，对斜风作用敏感；各风攻角下，颤振临界风速随着风偏角的增大均没有呈现如斜风分解法所述的余弦变化特征，其颤振稳定性可能在斜风下更差；斜风效应对颤振稳定性影响较小，静风效应则显著降低斜风下的颤振稳定性，总体平均降幅达到11.59%,在考虑斜风和静风作用的综合影响后，降幅更加显著，达到11.85%;虽然相较于双塔斜拉-悬吊协作体系桥，三塔斜拉-悬吊协作体系桥的结构刚度和斜风下的颤振稳定性均有所下降，但是比相同主跨的三塔悬索桥有明显提高，是一种大跨度多塔缆索承重桥梁适宜的结构型式。