【作者】 张超；

【导师】 刘宪林；

【作者基本信息】 郑州大学 ， 电力系统及其自动化， 2002， 硕士

【摘要】 配网作为电力网的末端，直接与用户连接，线路分布广，电压等级低，网上设备数量多。配网线损在整个电网线损中占有相当大的比例。挖掘配网的降损潜力，对提高供电企业的经济效益，具有重要的意义。 随着电力行业市场化逐步展开，电能损耗和经济效益将密切联系起来。研究简便、快捷、有效的线损分析方法将有利于寻找影响损耗的主要因素，掌握线损的变化趋势，改进线损管理工作，降低线损至最低限度，最终推动配网以至整个电网经济运行的实现。 线损分析工作是一项涉及面很宽的工作，它与线损计算、网络补偿、网络改造、运行方式以及线损管理工作紧密联系在一起，有些分析内容尚无成熟的模式可以借鉴，对已有的分析方法也存在着不少的争论，因此，如何实现有效的损耗分析还有待于进一步地研究。 随着计算机技术的发展，诸如MapInfo、GIS等先进软件已应用到电力系统之中。基于“网损分摊”原则并考虑过网线损的电网线损理论计算与分析系统开发软件已用到工程实际。采用BorlandC++与FORTRAN混合编程、基于Windows操作系统和C++ Builder4通用的客户／服务器开发等。针对单一电压等级线路的线损分析方法及可视化线损计算与分析软件还有待于进一步地研究。 线损分析的目标就是要将来自各方面的信息综合后给出一个最佳的分析结论，采集各类信息的目的也在于此，因此，研究如何有效利用各类可以采集到的信息，最大限度地提高分析的正确率，就显得非常有意义。目前的线损分析方法一般采用枚举法或遗传算法，但前者是一一进行尝试，工作量大，后者为解空间内的充分搜索，寻优过程时间长。鉴于一般的单子神经网和基于分解的组合神经网的不足，以及目前的线损分析方法的不足，本文采用基于信息类型和不同特征向量相组合的集成神经网进行分析。其基本思想就是通过对信息的有效融合，用各种子神经网从不同侧面进行分析，最大限度地提高分析的准确率。 本文提出并建立了基于集成神经网的针对配网特点和线损分析要求的配网线损分析数学模型。该模型包括分配层、融合层和输出层三层。采用模块化结构，融合技术贯穿其中，既可保证模块之间的联系又不失模块自身的相对独立性。模型参数的选取和确定可以通过样本学习、参考运行经验等方法获得，并可以随管理、运行或用户的要求进行动态修改。基于该模型的线损分析算法较为简便，速度较快，可以比较全面地反映各种因素对线损影响的程度，可为合理制定降损方案提供科学的依据。 开发了理论线损计算与分析软件。该软件采用图形化操作界面，数据库不仅独立运行而且还可以与具有类似结构的其他数据库相互共享数据或根据需要截取数据，根据理论计算数据对配网线损进行定量和定性的分析，分析结果以文本框、报表和图形的形式显示。软件操作简便，结果显示清晰，还可与其它程序相结合，避兔了编制其它软件需重新编译源代码，减少了程序源代码之间的相互依赖性，有利于软件的重构和扩充。 本文还通过配电线路实例对线损计算方法进行了对比分析，指出了运行时间选取对线损计算结果准确度的影响，并分析了其中可能的原因。 鉴于上述所做的工作只是初步的研究和探索阶段，因此，在以下几个方面有待于进一步地开展工作。 由于集成神经网建立在神经网和信息融合技术的基础之上，涉及到人工智能领域的大量知识。本文只是初步设计并探索了其模型的结构，如何将其应用到配网线损分析工程实际有待于进一步地研究。 线损分析系统为综合系统，在分析的过程中除了进行必要的线损计算外，还要根据所获得的来自各方的数据以及运行经验和相关的运行记录进行推理和判断，从而应将专家系统引入到线损分析中来。 设想将“损耗区域平面图”加入到线损分析中，通过“损耗区域平面图”可以描述配网中节点和支路对内部系统的关联及其对潮流分布的影响，发掘共有的特征和相异之处，为损耗分析提供可能的依据。 面向对象这种新的程序设计技术，按照不同对象的特性，把可能影响这个对象的方法和数据封装一起，按对象特性设计一个软模型，任何事件都将自动导致这个软模型按其特性反映出变化后的状况，而不问对象是否需要了解。可实现数据隐藏，提高软件开发与生产的效率和集成度，支持知识型和多媒体数据等，采用面向对象开发软件将是以后线损计算分析软件的趋势。 本文在查阅了大量的相关文献后，设计和提出了神经网和信息 融合技术相结合的针对配网特点的线损分析数学模型，并设计了与 线损相关的计算分析软件，经实例验证可取得较为满意的结果。 21世纪是知识经济时代，电力系统的发展应与时代特点相适 应。知识经济时代要求电力供应具有高可靠性和高质量。大力开展 电力信息化的研究和应用，有利于线损计算分析方法的完善，并最 终实现电力的可持续发展。更多还原

【Abstract】 As the terminal of an electric network, the distribution network is directly connected with the users. It is charactered with wide spread line, low voltage level, many devices on the line. Lineloss of distribution takes up considerable ratio in the general network. Picking out the potential is very important to the efficiency of supply enterprises.With the deployment of the marketization of electric power industries, loss will keep closely in touch with the benefit. The research of simple, quick, effective analysis is useful to find main factors effecting on loss, grasp the changing tendency of loss, improve the administration of line, reduce the loss into the minimum, and realize the economic operation of the whole electric power network in the end.The work of lineloss analysis involves much work which includes lineloss computation, network alteration, running style and administration, however some content has no model to reference, and existing analysis has much controversy. So how to proceed the efficient analysis of loss awaits research.With the development of technology in computer, advanced softwares such as Maplnfo, GIS and so on have applied into the electric power. Based on the principle of loss-allot considering the transnet loss, theoretic calculation and analysis system of lineloss is developed. And the software has been applied into the real engineer. The developed style runs to the hybrid program of Borland C++ and FORTRAN, and being built on the Windows and C++ Builder4 universal client/server. And the way of lineloss-analysis aiming at the single voltage level and visualized software about it keeps researching.The analyzed object of lineloss is to give an optimal analyzing conclusion by colligating the information from outers, which is also the meaning of the collecting all kinds of the information. So the research for effectively collecting and lifting the right ratio in the maximum seems very significant. Generally the present analyzing ways adopted are enumerating or genetic algorithm. The former tests all, which needs much work, while the latter is the complete ferret for space of resolution, which will take much time in finding the optimal. As for theshortcomings in the single neural net or synthetic neural net based on decomposition, and existing way, integrated neural network based on style of information and different eigenvector is adopted in this paper. Its basic thought is to effectively fuse the information, proceed the analysis from different aspects by variable children neural net, and heighten the accurate rate as possible.In this paper, a mathematical model of the lineloss-analysis of distribution, aiming at the characters of distribution and meeting the requirements of analysis is presented and established. There are three layers in this model, distribution, infusion and output. It adopts module-structure across with the technology of the infusion, which will ensure the relation between modules and respective relative independence. The selection and confirmation of parameters of modules can be obtained by the ways such as sample-studying, experience-referencing and so on and made dynamic adjustment according the requirements of users. The algorithm based on that module is simple, quick, can reflect the comprehensive extent of variable factors effecting on lineloss, and provide reducing program with scientific basis.A software of theoretical computation and analysis has been developed. In its design, DB can not only runs independently, but shares data with similar DB or intercepts data it needs. By the formed data the quantitative and qualitative analysis is performed. Then the results of analysis are shown in the form of text, report, and chart. And it operates simply, shows clearly, and also connects with others, which avoids re-editing the original codes for the designing others and reduces interdependence of codes, and promotes reconstruct and extension.By the example of distribution network the ways of calculation of lineloss has been made a comparable更多还原