电力科技图谱柔性构建方法

电力科技图谱应柔性应对新技术体系和新电力技术发展时必须面临的本体模型构建、存量数据标记、电力实体识别的影响.文章研究了一种电力科技图谱的柔性构建方法,首先,以科研项目为核心构建本体,向上关联技术体系,横向关联研究网络,向下关联技术实体;其次,提出将实体标记问题转为文本分类问题,采用主题分类和最大熵语义分类方法解决文献实...     

大规模电力系统复杂数据的通用匹配方法

针对当前数据匹配方法不能满足电力系统数据通用性的问题,考虑电气量的属性相似度建立了基于反馈机制的多指标相似度模型。引入基于余弦算法的文本相似度计算方法,对其根据电力系统特征词的语义模型进行改进。采用分层逐级匹配和基于拓扑的数据匹配策略,给出了计算文本相似度与属性相似度权重的动态计算方法。提出的方法已经成功应用于某省级电网公司整定计算专家系统,以及能量管理系统和故障信息系统与整定计算系统的数据匹配,验证了该方法的通用性与准确性。     

六种基于能量的潜在蒸散发估算方法在海河流域的适用性

潜在蒸散发是一个综合性气候因子,是流域水文循环模拟的主要输出量。针对现存潜在蒸散发量估算方法的适用性分析不够全面的问题,基于海河流域45个国家气象台站1960~2010年日气象数据序列,以Penman-Monteith方法估算量和蒸发皿蒸发量为基准,从潜在蒸散发估算量与蒸发皿蒸发量的相关性、年均值空间变化特征、年内四季空间变化特征和年际变化趋势四个方面,探讨了六种基于能量的潜在蒸散发估算方法的适用性。结果表明,Penman-Monteith方法在整个流域具有较强的适用性;六种基于能量方法中,Doorenbos-Pruitt方法、Makkink方法、Hargreaves方法估算效果较好。     

基于文本分类的输变电设备缺陷分析

为提高电力设备缺陷文本利用效率,构建缺陷文本分类模型。首先分析了中文文本分类的流程,然后结合电力设备缺陷文本的特点,构建了4层卷积神经网络的电力缺陷文本分类模型;最后以某市近10年来的电力设备缺陷文本记录作为数据来源,结合上述的模型,对数据进行训练和分类,并与部分传统的机器学习分类模型进行比较。结果表明,提出的分类模型的错误率为2.86%,远低于传统的6.99%,具有明显的优势。     

基于流形学习的汽轮机转子振动故障特征提取

流形学习算法是一种非线性的数据降维方法,通过对高维空间数据的学习,可以获得数据的低维几何结构,比传统的维数约简方法更能体现事物的本质。针对汽轮机转子故障诊断中原始特征并不具有良好的可分性,必须进行特征提取这一问题,提出了将流形学习理论应用于汽轮机转子振动故障特征的提取的方法,极大地保留原始特征的非线性信息,改善了故障诊断的分类识别能力。实验证明,基于流形学习的汽轮机转子振动故障特征提取方法,提高了故障诊断的准确率。     

火电厂振动数据的二维提升小波压缩算法

为了在数据压缩过程中更有效地降低振动数据中存在的冗余,提出了一种新的火电厂振动数据压缩与解压缩算法.首先将待压缩的实时数据从一维空间转换到二维空间,然后采用基于二维提升格式的对称双正交小波滤波器组对数据进行压缩和解压缩.利用实际测取的电厂振动数据对算法进行了验证.结果表明,在相同的压缩比下,基于二维小波压缩的效果优于基于一维小波压缩的效果,因此该算法是一种有效和实用的振动数据压缩方法.     

负荷模型辨识中广域电网负荷空间分类

提出了负荷模型辨识中广域电网负荷的空间分类方法,基于工业、商业、农业、居民及其他负荷的典型值数据,通过模糊均值聚类（FCM）算法对负荷按负荷性质空间分类,辅以灵敏度计算公式确定重点辨识参数,进而以遗传优化算法并结合暂态过程各种扰动设置中电压响应曲线的交互计算,以全网母线电压跌落最为严重的母线作为观察变量,辨识修正负荷模型参数,并以不分类、区域分类两种方法与文中所提出方法作对比,仿真结果表明,按负荷性质分类具有合理性与有效性。     

基于Bert的电网故障处置预案信息抽取研究与实现

电网故障处置预案是及时、正确处置故障的重要依据。对其关键信息进行抽取,是实现计算机理解预案内容并进一步支撑故障处置智能化的重要基础。传统的信息抽取方法多为模式匹配,其特征和模式的获取需要专家经验,费时、费力且泛化能力较差。为实现电网故障处置预案文本信息的高效、自动抽取,文章提出一种基于Bert(bidirectional encoder representations from transformer)模型的序列标注方法。具体如下：(1)提出一种针对处置预案文本特点的数据预处理方法,该方法将句子分为不同的类别;(2)基于Bert模型自动学习和获取文本特征,并使用标注数据微调模型的方法进行信息抽取;(3)对使用上述分类方法处理过的预案文本,提出一种对每个类别分别进行模型训练并获得序列标注结果的方法。多组对比实验结果表明,该方法可以有效提升信息抽取的准确率。     

基于空间金字塔模型的DCA特征融合地形分类

针对空间金字塔视觉词袋模型对地形图像分类时忽略颜色信息、纹理表达不明显及特征维度高等问题，提出一种基于空间金字塔模型的DCA特征融合地形分类方法。该方法优化传统空间金字塔模型子区域划分方式，提取地形图像优化后的SPM-BOVW特征、HSV特征、LBP特征；通过DCA算法构建3组变换特征；采用串联将变换特征进行融合。实验结果表明，以融合特征作为支持向量机（SVM）分类器的输入，利用网格参数寻优，最终获得了较高的地形分类精度，说明所提方法在太阳能电站的地形图像分类上具有较好的鲁棒性。     

基于改进TFID F特征加权算法的 95598投诉工单分类实现

分析了传统朴素贝叶斯算法、TFIDF特征加权算法在文本分类建模应用上的缺陷,在此基础上提出了TFIDF的优化算法,并分别采用传统朴素贝叶斯算法、基于TFIDF的朴素贝叶斯算法、基于改进TFIDF的朴素贝叶斯算法分别对某供电公司投诉工单进行了分类建模。研究结果表明,改进TFIDF算法能在TFIDF算法的基础上进一步提升分类器的性能,投诉工单分类准确率达到93.03%,较传统朴素贝叶斯算法提升了4个百分点。     

基于多层深度互信息变分网络的风电机组轴承故障预警方法

针对变工况复杂环境下风电机组轴承的早期故障潜隐性高且故障阈值设置困难的问题,提出一种基于多层深度互信息变分网络的轴承故障超前预警方法。该网络在变分自编码器的架构上进行多层编码拓展,采用解码信号的二次编码增强了变分网络对输入信号中噪声的鲁棒性,通过隐层变量与输入信号间最大化互信息以及与二次编码特征间的最大化互信息,提高了网络对轴承正常状态空间的建模能力。以二次编码特征与隐层变量的重构误差为基础,构建健康指数,结合三次递推指数加权移动平均模型设置与风电机组轴承工况环境相适应的健康基线。在2个风电机组轴承故障检测试验中的应用结果表明,该方法对比传统的基于模型重构的轴承早期故障检测方法,具有更高故障预警准确率和抗干扰能力。     

A new importance assessment method for risk‐informed SSC categorization

This paper proposed a new importance assessment method and new importance measure to perform risk‐informed system, structure, and components (SSCs) categorization task, regarding the dynamic of SSCs' importance measures caused by varied configurations in nuclear power plants (NPPs). The new method consists of 3 importance measures, Fussell Vesely, risk achievement worth, and return to service worth, and an integrated time ratio factor is introduced to get an SSCs ranking, which could handle more realistic configurations in NPP operation. The proposed method has been used in CANDU feedwater system SSCs categorization, supported by RiskA and RiskAngel software, and a different SSCs categorization result compared with NEI method was given. The difference of categorization result derived from 2 methods was discussed in this paper, and the effectiveness has been demonstrated by the NPP's real operation statistics.     

Design of isolated renewable hybrid power systems

Isolated electrical power generating units can be used as an economically viable alternative to electrify remote villages where grid extension is not feasible. One of the options for building isolated power systems is by hybridizing renewable power sources like wind, solar, micro-hydro, etc. along with appropriate energy storage. A method to optimally size and to evaluate the cost of energy produced by a renewable hybrid system is proposed in this paper. The proposed method, which is based on the design space approach, can be used to determine the conditions for which hybridization of the system is cost effective. The simple and novel methodology, proposed in this paper, is based on the principles of process integration. It finds the minimum battery capacity when the availability and ratings of various renewable resources as well as load demand are known. The battery sizing methodology is used to determine the sizing curve and thereby the feasible design space for the entire system. Chance constrained programming approach is used to account for the stochastic nature of the renewable energy resources and to arrive at the design space. The optimal system configuration in the entire design space is selected based on the lowest cost of energy, subject to a specified reliability criterion. The effects of variation of the specified system reliability and the coefficient of correlation between renewable sources on the design space, as well as the optimum configuration are also studied in this paper. The proposed method is demonstrated by designing an isolated power system for an Indian village utilizing wind-solar photovoltaic-battery system.     

基于多层语义融合注意力机制的短期风电功率概率密度预测方法

获得未来风电功率的短期概率性信息将有助于电网的综合能源调度,为此提出一种结合多层语义融合注意力机制的短期风电功率概率密度预测方法。为尽可能获得更多语义层下的编码信息,在编码阶段引入循环高速网络（RHN）并通过深层的RHN网络结构最大程度地提取输入特征的底层关联信息。设计多层语义融合注意力机制以融合不同语义层下的局部注意力向量,进一步加强编码特征向量的表达能力,并将网络的输出与分位数回归和核密度估计方法结合,得到不同分位点下未来短期风电功率的预测结果与连续概率密度分布。实验结果表明：提出的短期风电功率概率密度预测方法不论是在预测的得精度上,还是在具有不确定性的预测结果分布上均优于其他比较模型。     

基于改进Mask-RCNN模型的电力设备锈蚀识别

针对变电站电力设备锈蚀检测问题,文章提出了一种基于改进Mask-RCNN模型的锈蚀检测方法.使用残差网络Resnet101作为模型的基础网络,使用改进后的非极大值抑制(Non-Maximum Suppression,NMS)算法,提升语义分割的检测准确率.针对设备锈蚀数据集,使用dropout算法对Mask-RCNN提...     

A fixed grid numerical modelling methodology for convection-diffusion mushy region phase-change problems

An enthalpy formulation based fixed grid methodology is developed for the numerical solution of convection-diffusion controlled mushy region phase-change problems. The basic feature of the proposed method lies in the representation of the latent heat of evolution, and of the flow in the solid-liquid mushy zone, by suitably chosen sources. There is complete freedom within the methodology for the definition of such sources so that a variety of phase-change situations can be modelled. A test problem of freezing in a thermal cavity under natural convection is used to demonstrate an application of the method.     

Latent heat storage for solar energy systems: Transient simulation of refrigerant storage

This paper presents a brief review of the available latent heat storage systems for solar energy utilization. A new concept of latent heat storage of solar energy via the refrigerant-absorbent mass storage in absorption cycle heat pump systems used for solar space heating/cooling has been proposed and assessed thermodynamically. A computer modelling and numerical simulation study shows that the concept of refrigerant storage is fundamentally sound, technically feasible and yields the following advantages over other storage methods: (i) the storage capacity per unit volume is high as the latent heat of vaporization of the refrigerant is high; (ii) the heat loss from the storage to the surroundings is minimum as the storage temperature is near the ambient; (iii) prolonged energy storage is possible with no degradation in system performance and hence suitable for combined solar heating and airconditioning. The effects of operating parameters on the energy storage concentration and storage efficiency have been studied in detail.     

ERAL SOURCE-BASED METHOD FOR SOLIDIFICATION PHASE CHANGE

After a brief review of current source-based methods for modeling solidification phase change systems, a new source-based method for the treatment of latent heat evolution is presented. The essential feature of the proposed method is linearization of the discretized source term. This results in a robust and accurate computational method that can deal efficiently with a wide range of latent heat evolution mechanisms (i.e., liquid fraction temperature relationships). The proposed method is illustrated on application to a test problem in which various liquid fraction temperature relationships are employed.     

传统新安江模型在流域分布水文模拟中的应用

结合传统新安江模型的特点,基于DEM数据,借助GIS空间分析功能,提出了采用流域—产汇流分区—单元面积的流域空间离散法,给出了模型参数空间化的方法,以期使该模型更好地适应流域分布式水文模拟。