【作者】 张健；

【导师】 朱林； 苏忠阳；

【作者基本信息】 华南理工大学 ， 电气工程（专业学位）， 2021， 硕士

【摘要】 在十四五规划和“碳中和”战略的引领下,清洁能源发电比重将会保持逐年提升。针对含大规模新能源的电力系统,对其进行完整的电磁暂态仿真建模耗时费力,而采用动态等值技术有效地化简电力系统规模,这是对大系统进行安全稳定分析行之有效的方法。发电机聚合等值是动态等值的核心环节,狭义是指同步发电机进行同调分组后再进行参数聚合,从而获得等值机模型及其参数。然而,在风电得到迅猛发展的背景下,传统的以同步发电机为核心的等值也迫切需要增加新的内容,从而适应电源侧的发展趋势。数据驱动理念为解决发电机等值提供了新思路,即不再依赖于建立详细的物理模型,通过数据挖掘,利用深度学习算法和数据分析方法构建等值模型来逼近真实模型。基于上述理念,本文拓展了传统发电机等值方法框架,从分组和参数聚合两个环节展开同步发电机和风力机组的等值研究。本文基于上述研究思路,主要完成了下述工作:(一)提出了基于数据驱动的同步发电机三阶段分群方法。首先,充分挖掘历史运行数据获取实际电网拓扑信息,快速实现发电机同调预分组。然后提出了基于卷积神经网络的发电机二次分组方法,以故障下机端信息为输入特征,来训练CNN分组模型。训练后的CNN模型可用于预测发电机分组。最后利用K-means算法将剩余发电机进行聚类。所得的方法提高了实际电网中同步发电机同调分组效率与准确性,合理的分组结果为后续发电机参数聚合及动态等值的网络化简打下了坚实基础。(二)提出了基于时域的电磁回路参数辨识方法及控制器交替寻优方法。本文在上述分组的基础上对等值机电磁回路提出了基于时域的参数辨识方法,以边界节点的输入输出一致性为辨识目标,该方法能较好地维持边界节点特性,满足动态等值要求。同时,本文对等值机控制器参数聚合提出了灵敏参数与非灵敏参数交替寻优的聚合方法。先对灵敏参数利用改进的粒子群优化算法进行寻优,再对非灵敏参数进行寻优。所得控制器参数较全参数频域寻优有更高的精度,更好的保持了原系统的动态特性。(三)提出了基于等效功角和功率权重的双馈风电场等值方法。本文对双馈风电场的分群提出了基于等效功角的数据分析方法,以等效功角的相似度作为分群依据。等效功角能表征双馈风力发电机的动态特性,可用作同调判据。提出了基于Prony算法来量化各风电场的等效功角数值并用于分组。最后对等值风机参数聚合提出了基于功率加权的算法,能有效地保持大机组动态特性,与原系统特性更为贴近。最后,以实际南方电网数据为例,运用上述方法对电力系统中所有同步发电机及风电场完成等值,与传统发电机等值方法进行了对比验证了本文所提方法的有效性和准确性,解决了同步发电机、风力发电机的分组和参数聚合工作中严重依赖物理模型、耗时较长及精度不高等问题,支撑了实际交直流电力系统的动态等值要求。更多还原

【Abstract】 Under the guidance of the 14 th Five-Year Plan and the "carbon neutral" strategy,the proportion of clean energy power generation will continue to increase year by year.For power systems with large-scale new energy sources,it is time-consuming to carry out complete electromagnetic transient simulation modeling.Dynamic equivalence technology is used to effectively simplify the scale of the power system and reduce the number of generators,which is an effective method for safety and stability analysis of large systems.Generator aggregation is the core steps of power system dynamic equivalence.It refers to synchronous generator grouping and then parameter aggregation,so as to obtain parameters of the equivalence model.However,in the context of the rapid development of wind power,the traditional equivalent work with synchronous generators as the core also urgently needs to add new content to adapt to the development trend of the power supply side.The data-driven concept provides a new idea for solving generator equivalence,that is,there is no need to establish a detailed physical model,through data mining,the use of deep learning algorithms and data analysis methods to build an equivalent model to approximate the real model.Based on the above concepts,this article expands the framework of the traditional generator equivalence method,and starts the equivalence study of synchronous generators and wind turbines from two steps: grouping and parameter aggregation.For synchronous generators,considering the use of deep learning algorithms to distinguish the coherence of generators based on the data under disturbance.Parameter aggregation aims at keeping the node data consistent,and continuously fitting parameters of the equivalence model in the time domain.For wind turbines,physical quantities that can characterize their dynamic characteristics are derived,and their similarities are quantified by data analysis methods and used for grouping.Equivalent wind turbine models aggregate parameters with output power as the weight.Compared with the traditional method of generator aggregation and equivalent,the data-driven method proposed in this article has advantages in speed,repeatability and accuracy.Based on the above research ideas,this article mainly completed the following tasks:(1)A data-driven three-stage clustering method for synchronous generators is proposed.First,using the historical data to obtain practical power grid topology and quickly complete generator pre-grouping based on constraints such as geographic location and separation of thermal power and hydropower generators.Then,a method for secondary grouping of generators based on Convolutional Neural Network(CNN)is proposed,which takes pregrouped coherent generator fault information as input features and group number as a label to train the CNN model.After training,the CNN model is used to predict the groups of generators.If the threshold requirement is met,it will be included in the existing group.Finally,the remaining generators are clustered using the K-means algorithm,and the Gap statistic algorithm is used to determine the optimal number of clusters.The obtained reasonable coherent grouping results lay a solid foundation for subsequent parameters aggregation and network simplification.(2)A time-domain-based electromagnetic loop parameters identification method and the controller alternate optimization method are proposed.This paper proposes a parameter identification method based on the time domain of the equivalent machine based on the above grouping.The input and output consistency of the boundary node is the identification target.This method can better maintain the characteristics of boundary nodes and meet the requirements of dynamic equivalence.At the same time,this paper proposes an aggregation method of alternately optimizing sensitive parameters and nonsensitive parameters of the equivalent generator controller.First,the trajectory sensitivity is used to distinguish the sensitive parameters and the non-sensitive parameters.The sensitive parameters are optimized using the improved particle swarm optimization algorithm with the dominant generator parameters as the initial value,and then the non-sensitive parameters are optimized.The obtained controller parameters have higher accuracy and better maintain the dynamic characteristics of the original system.(3)The equivalent method of doubly-fed wind farm based on equivalent power angle and power weight is proposed.This paper proposes a data analysis method based on Equivalent Power Angle(EPA)for the grouping of doubly-fed wind farms,and uses the similarity of EPA as the basis for grouping.EPA is derived from the physical model of the doubly-fed wind turbine,which can reflect the operating characteristics of different operating conditions under various control modes,that is,it can characterize the dynamic characteristics of the doubly-fed wind turbine and can be used as a criterion for coherence.A data analysis method based on the Prony algorithm is proposed to measure the EPA value of each wind farm.When the EPA similarity between the wind farms meets the threshold requirement,the equivalent aggregation can be performed.Finally,an algorithm based on power weighting is proposed for the aggregation of equivalent wind turbine parameters,which can effectively maintain the dynamic characteristics of large wind farm and is closer to the original system characteristics.Finally,using the actual China Southern Power Grid data as an example,the above method is used to complete the equivalence of all synchronous generators and wind farms in the power system.Compared with the traditional generator equivalence method,the methods in this paper are effective and accurate.This paper introduces the concept of data-driven and carries out research on generator aggregation methods.It solves the problems of reliance on physical models,long time-consuming and low accuracy in generator coherent grouping and parameter aggregation,which satisfies the requirements of dynamic equivalence in practical AC and DC power systems.更多还原