【作者】 李培强；

【导师】 李欣然；

【作者基本信息】 湖南大学 ， 电气工程， 2009， 博士

【摘要】 综合负荷模型是决定仿真研究结果可信赖程度的关键因素之一。因此负荷建模是电力系统重要的基础性课题。由于综合负荷自身的随机时变性、地域分散性、成分多样性、严重非线性等特点,建立完全精确的数学模型是十分困难的工作。正是由于负荷模型的重要性和复杂性,负荷建模正成为电力系统仿真分析领域的具有挑战性的研究方向,引起了电力工程界与学术界的广泛关注。论文围绕静态模型参数辨识和行业特性分析、建模用户精选和变电站特性综合、实测建模数据特征提取、神经网络负荷模型及其综合能力开展研究,主要内容包括:1、在分析传统优化算法应用于负荷模型参数辨识的主要缺陷的基础上,提出了基于二次规划的Lemke优化算法的负荷静态模型参数辨识的新方法。通过对电力负荷元件及其组合进行的静态故障模拟,对实验故障数据进行负荷建模和参数辨识,辨识结果证实了该方法的正确性和有效性。基于统计综合法负荷建模的基本原理,结合实际系统的建模实践,提出了一套基于统计综合法负荷建模的系统方法,通过统计综合的方法从而确定行业的综合负荷幂函数特性。2、基于统计综合法负荷建模的基本原理,提出在统计综合法负荷建模的第一个层次,即用户特性调查阶段,对于用电行业的典型用户,在初步专家筛选的前提下进行负荷结构特性调查,在对调查数据进行分析处理的基础上,提出了基于模糊等价关系和模糊C均值算法的两种分类方法进行行业典型用户的精选,为准确筛选行业典型用户从而建立各行业的综合负荷特性提供了新的方法。通过纺织行业用户精选为例,论证该方法的可行性和有效性。3、对湖南电网的变电站负荷特性进行了分类与综合。其思路是在对变电站静态负荷构成特性参数标准化的基础上,以此为负荷特性分类和综合的特征向量,应用所提的两种分类算法,对某电网枢纽变电站进行了聚类分析。在负荷特性调查数据的基础上,利用综合处理的方法得出了湖南电网48个220kV变电站3种不同运行方式下感应电动机负荷和恒定阻抗负荷的实际比例。4、根据随机理论思想,把负荷建模的实测输出数据看成随机扰动电压的响应,基于小波包的分解和重构理论,对实测建模电流信号进行3层小波包分解,用Wprcoef函数对小波分解系数进行重构,准确提取和构造了负荷实测扰动建模数据的能量特征向量。在特征向量归一化的基础上,利用减法聚类的模糊算法对特征向量进行分类处理,获得了理想的分类结果。5、提出一种基于减法聚类的模糊神经网络的负荷建模新方法。首先对建模样本输入输出数据进行特征分析,应用减法聚类自适应的调整建模数据的聚类数和聚类中心,以确定负荷模型的模糊规则数和隶属度函数个数。通过神经网络对推理数据进行学习,用反向传播算法来修正网络的连接权重,辨识模糊模型的隶属函数的参数,完成综合负荷的非机理建模。在此基础上,对模糊神经网络负荷模型的综合能力进行验证。更多还原

【Abstract】 Synthesis load modeling has been received widespread attention in power system analysis and control. The paper has explained static model structure and basic identified theory about parameter of load component of electric power. A new Lemke optimal algorithm is applied in identified parameter of the load component. It make up static simulation experiment to different loads component in the paper, using characteristic recording instrument of load, in which can write down its electric voltage and current. The modeling practice shows its validity and feasibility of the method.In the electrical power system simulation computation, the synthesis load model of induction motor and the constant impedance proportion has the important influence to the simulation result extremely. The paper based on the statistical synthesis load modeling mentality, has carried on the detail investigation statistics to the Hunan electrical network load characteristic. In the load characteristic investigation foundation, the paper has obtained the actual proportion of induction motor and the constant impedance load of 48 transformer substation under 3 different operational condition Hunan electrical grid using the synthesis processing method, the conclusion has the vital significance to the electrical power system simulation computation.The paper have put forward a set of system approach of the component-based modeling Approach. On the premise of analysis and procession the data, the paper has used two fuzzy clustering to choose the choice industry consumers, that is, the fuzzy equivalent relation clustering and the fuzzy C means clustering. The paper is analyzed 48 transformers in the two methods and the class synthesis characteristic of the fuzzy equivalent relation is obtained using the weighted average method. The fuzzy C means clustering results is more reasonable and effective than the fuzzy equivalent relation clustering. Two clustering methods have solved the complexity and subjectivity of the transformer load characteristic classification in load modeling.Load measured data is regarded as stochastic disturbance of voltage. The paper applies three layer decomposition and re-construction of wavelet package method to analysis load modeling data, attain characterizes vector of load data, construct characteristics vector, which is used to classify load data. Based on the characteristics vector is standardized, load data is classified by fuzzy subtract clustering in the paper. The method is proved valid, which is high precision and convergence by the example of the attaining characteristics and clustering of dynamic lab and transformer substation data. It is the significant to load modeling processing the amount of load data.To obtain accurate load modeling, a new load modeling method is proposed in the paper, which is fuzzy neural network load modeling based-on Subtractive clustering. By analyzing input and output data, the paper set up the mount function to classify its clustering number and adjust class center. In this way, the method confirms member function parameters of the initial fuzzy load model. The method can obtain fuzzy rules through the neural network study to modeling data, optimize the parameter member function by amending the linkage proportion in the back prevalence algorithm. So it can identify the load model construct and attain optimal parameter of function member. The load model synthesis ability is proved by stimulating other load data.更多还原