Distributed Optimal Power Flow Using ADMM
- 格式:pdf
- 大小:2.19 MB
- 文档页数:11
基于自适应步长ADMM的直流配电网分布式最优潮流韩禹歆;陈来军;王召健;刘炜;梅生伟【摘要】Direct current distribution network has broad prospects,and its optimal power flow(OPF)problem has important significance to economical operation.Following the advances in second-order coneprogramming(SOCP)convex optimization,this paper establishes the SOCP-OPF model in which the constraints of voltages,currents and power are considered for direct current distribution network in redial topology.In addition,a distributed OPF calculating method based on alternating direction method of multipliers(ADMM)is proposed to overcome many difficulties faced by traditional centralized dispatch paredwith existed researches,each bus is equipped with a computing agent and the agents exchange simple information with neighbors in every iteration,then solve the optimization problem in a parallel pattern without centralized global coordination.The SOCP-OPF model takes the current limits on the lines into consideration,which makes the model more practical.The proposed method also has a dynamic step size adjusting strategy to improve the efficiency of calculating.Numerical simulations on typical system show the good convergence and accuracy of the algorithm.%直流配电网的发展前景广阔,其最优潮流(OPF)问题关系到电网经济运行,具有重要的工程意义.针对放射状直流配电网,以二阶锥规划(SOCP)凸松弛理论为基础,建立了考虑电压、电流、功率约束的SOCP-OPF凸规划模型,并提出一种基于交替方向乘子法(ADMM)的分布式最优潮流计算方法,以解决传统集中式优化方式面临的诸多难题.相比已有研究,该方法在各节点配置计算单元,无需全局协调或分层分区,利用相邻主体间少量的信息传递即可通过并行计算得出全局最优解;优化模型中考虑了配电线路传输电流限制,约束条件更全面;计算方法中设计了自适应步长调整机制,计算效率较高.IEEE 33节点和IEEE 123节点的算例分析验证了所提算法的准确性和良好的收敛性.【期刊名称】《电工技术学报》【年(卷),期】2017(032)011【总页数】12页(P26-37)【关键词】直流配电网;分布式优化;最优潮流;交替方向乘子法;自适应步长【作者】韩禹歆;陈来军;王召健;刘炜;梅生伟【作者单位】电力系统及发电设备控制和国家重点实验室(清华大学电机系) 北京100084;电力系统及发电设备控制和国家重点实验室(清华大学电机系) 北京100084;陕西省地方电力集团有限公司西安 710061;陕西省地方电力集团有限公司西安 710061;电力系统及发电设备控制和国家重点实验室(清华大学电机系) 北京 100084【正文语种】中文【中图分类】TM744随着分布式发电(Distributed Generation,DG)技术及电力电子技术的发展,直流配电网已在诸多方面具备一定的经济与技术优势。
高三环保行动英语阅读理解30题1<背景文章>Environmental issues have become one of the most pressing concerns in the world today. Climate change is a major threat, with rising temperatures leading to melting ice caps, rising sea levels, and more extreme weather events. Pollution is also a significant problem, affecting the air we breathe, the water we drink, and the land we live on.Air pollution is caused by a variety of factors, including industrial emissions, vehicle exhaust, and burning of fossil fuels. This can lead to respiratory problems, heart disease, and even cancer. Water pollution comes from industrial waste, agricultural runoff, and sewage. It can contaminate drinking water sources and harm aquatic life. Land pollution is caused by improper disposal of waste, such as plastic, and can damage soil quality and wildlife habitats.In addition to these problems, deforestation is also a major issue. Trees play a crucial role in absorbing carbon dioxide and producing oxygen. When forests are cut down, it releases large amounts of carbon dioxide into the atmosphere, contributing to climate change.We must take action to address these environmental problems. This can include reducing our carbon footprint by using renewable energysources, reducing waste, and protecting natural habitats. We can also support policies and initiatives that aim to combat climate change and pollution.1. What is one of the major threats caused by climate change?A. Decreasing temperatures.B. Stable sea levels.C. Melting ice caps.D. Less extreme weather events.答案:C。
Research on Smart Grid in China Jingjing Lu, Da Xie,Member, IEEEand Qian Ai,Member, IEEEAbstract--The Smart Grid is the latest direction for the futurepower system development. In this paper, firstly the backgroundof Smart Grid, its meaning, as well as the concept and structurewere presented. Typical diagram of Smart Grid was illustrated.Then, the current development of Smart Grid in United States and Europe were described, development ideas and the future trends in these countries were summarized and compared as well.Besides, the driving force of Smart Grid in China was analyzed,with detailed introduction of current related projects in China. The relation between the UHV Power Grid and the Smart Grid was discussed. Finally, the potential role of Smart Grid in future power grids in China was prospected and a new direction for China’s Smart Grid development was charted.Index Terms—Smart Grid, UHV power grid, planning,operation, managementI. INTRODUCTIONWith the promotion of world economy modernization, theprice of oil has been kept on a upward trend. What is also noticeable is the shortage of energy supply around the world, the increasing pressures on resources and environment pressure, and the enormous power losses in energy delivery due to the low efficiency of the current power grid. What’s more, owing to the growing electricity demands and the users’increasing requirements for reliability and quality, the power industry is now facing unprecedented challenges and opportunities. Therefore, a new sort of power system of environment friendly, economic, high performance, low investment, safety, reliability and flexibility has been a goal of engineers in power industry.Still, the emergence of advanced meter infrastructure and more extensive usage of the Internet accelerate the process [1]. Since 1990's, with the increasing use of distributed generation power, more demands and requirements have been proposed for power grid intensity [2], [3]. To find out a optimal solution for these problems, power companies should accept the idea of new technology adoption, potential mining of the existing power system and improvement of its application and utilization. Consensus has been reached by experts and scholars from different countries that future power gird must be able to meet various requirements of energy generating and the demands of highly market-oriented power transaction so that the needs of the self-selection from customers can be satisfied individually. All of these will become the future development direction of Smart Grid.This paper focuses on the status of the development of the Smart Grid, analyzing the driving force of the Smart Grid and introducing the current demonstration projects in China. It also discusses the relation between UHV power grid and Smart Grid, and then prospect the significance of Smart Grid in the future. A new direction for Chinese Smart Grid development is charted as well, which might be the reference for the development of Smart Grid in China.II. CONCEPT OF SMART GRIDSmart Grid is a gradual development process accompanied with the technology innovation, demands of energy saving and managements needs. People will have their own understanding for Smart Grid, no matter if they are facility suppliers, IT companies, consulting firms, public power companies or power generation companies. From the earlier smart intelligence metering toelectrical intelligence, from transmission and distribution automation to a whole intelligent process, the concept of smart power grid has been enriched substantially [4]. In 2006, US IBM presented a "Smart Grid" solution. This is a relatively complete concept for current Smart Grid which indicates its official birth [5].As shown in Fig.1, a Smart Grid is basically overlaying the physical power system with an information system which links a variety of equipments and assets together with sensors to form a customer service platform. It allows the utility and consumers to constantly monitor and adjust electricity use. The management of operation will be more intelligent and scientific based on the dynamic analysis of needs both from user-side and demand side which can increase capital investment efficiency due to tighter design limits and optimized use of grid assets.In comparison with traditional grid, Smart Grid includes integrated communication systems, advanced Sensing,metering, measurement infrastructure, complete decision support and human interfaces.III. CURRENT RESEARCH ACTIVITIESA. Comparison of researches in Smart Grid area between European and the United statesIn the United States, there were several large power outages in recent years. Because of which, electric power industry pays closer attention to power quality and reliability;customers draw out more requests for electricity supply. The ever-increasing demands of national security and environmental protection policy of the United States leads to the establishment of a higher standard for power grid construction and management [6]. At the same time, in recent years’researches of basic materials, power and information technologies, breakthroughs have been achieved for implementation target which shows the significant improvement of reliability, efficiency in power network. Such as the emergence of superconducting cables, it assures Obama’s new government of United States has seen the daylights of the Smart Grid.Similarly, the European power users also raise higher requirements for electricity supply & power quality [7].Because of the extreme attention for environmental protection,compared with the construction of power grid in US,Europeans have more concerns about the construction of renewable energy access, the impact on wildlife, as well as the actively research on real-time monitoring and remotecontrolling.All is about to realize the "Plug & Play use" idea,ensuring a more friendly, flexible access and interaction with the user. In both Europe and United States, the most common direction for grid development is to seek new and renewable sources for energy generation. However, Smart Grid is not a fixed, static project, according to their particular status and main problems, all countries need to simplify the Smart Grid and make it adjusted to fit their own features.B. Driving force of the Smart Grid in ChinaThe drivers for Smart Grid construction can be concluded into market, circumstances, safety and power quality. Chinese power industry is also facing the similar situation as in Europe and the United States.At Market-oriented reforms level, the national network and unified national electricity market has not completely formed.In national wide, power exchanges are not effective; neither does the true meaning of the online bidding. From a long term view,China's transaction approaches of power markets and pricing structure is developing, market demand and supply sides will have more frequent interactions. In order to attract more users to join the market competition, power companies must improve their service, strengthen the interaction with users and provide more products for selection, so as to meet the demands of different types of users.At the macro policy level, the power industry needs to meet the requirements of resource-saving and environment-friendly society’s construction, adapt to climate change and suitable for sustainable development.Regarding the Chinese power grid itself, a strong backbone network has not been built yet, and it is still not strong enough to withstand multiple faults circumstances. The regional power grid backbone is also in a lower stability level, which results in a limited flexibility for system operation, etc. The snow storm weather in early 2008 which led to a blackout in major area of China vividly exposed the weakness of the current Chinese grid in safeguard of electricity supply aspect.Moreover, the lack of intelligent power distribution leads to a regional, seasonal shortage of electricity and coexistent in some areas with both surplus & shortages of electricity.There still remain challenges that how to improve the efficiency of power investment and construction, how to ensure the security and reliability of power grid’s operation, to ensure power quality; how to improve the maintenance of power system; how to enhance the service quality to users, as well as how to improve the power grid management in China.For these issues, Smart Grid would be an ideal solution.C. Current research activities in ChinaIn the year of 2006, IBM published the guideline named ‘Establishing Smart Power Grid and Innovating Management Methods –A New Thought of the Development of Electric in China’. Directed at the current opportunities and challenges of the power grid companies in China, the guideline suggests improving the efficiency of electrical investments and construction, the stability of power grid, and the companies’service and management level through the construction of smart power grid and the innovation of management methods.Meanwhile, IBM proposed that it can provide a whole scheme - Solution Architecture for Energy (SAFT) for the power companies in China to use the smart power grid effectively. SAFT contains several parts: first is to improve the digital level by connecting the equipments with sensors;Second is to establish the data collecting and integrating system; Third is to analyze: SAFT optimize the operating progress and management based on the analyzing of the data[5]. This is the bud of smart power grid in Chin.In October 2007, East China Power Grid Company embarked on the research area of the feasibility of smart power grid. The research project was not only correlated with the progress of those advanced companies and research facilities abroad, but also take the current situation and future needs of east china power grid into consideration. The result came out as, based on the high equipment level and strong technological innovation ability, the construction of smart power grid is feasible in east china power grid. East China Power Grid Company would follow the belief that‘Concerning the future and change fast with needs, and providing high quality service’, when building smart power grid. There is a three step strategy with an advanced power grid distributing center built by 2010, the construction of digital power grid with primary intelligence completed by 2020, and a smart power grid with the ability of self-healing built by 2030 [8]. The construction plan is still under consideration.On Feb. 28, 2009, as a part of the smart power grid of East China Power Grid Company, the three-state security defense and power generation monitoring system passed the acceptance check in Beijing, which stands for stable-state,transient-state and dynamic-state. The system integrated three single systems altogether for the first time, which includes power management system, power grid dynamic wan monitor system and online stability analysis and warning system. The operator has full access to the whole view of the power grid operating situation and the decision-making assistance without switching in systems or platforms. Besides, the system can effectively improve the management standardization and the level of flow of the related power plants through establishing the management checking platform and the assistance marketservice quality analyzing platform.The development of smart power grid research in China is slow and far behind the west. So far, only East China Power Grid Company and North China Power Grid Company have carried out researches about the developing and implementation plan. It is a tradition for China to emphasize technology development, and in fact, the equipments in China are more advanced than those in developed countries. Thus,smart power grid has a bright prospect in China.IV. PROSPECTS OF SMART GRID IN CHINAIn order to solve the problems of imbalance distribution for generation resources and power loads, the transmission capacity should be enhanced by building long-distance and large-capacity power transmission systems. And unity or united UHV power grids should be constructed under coordinated plan. The transmission of power on a large scale from west and north China to middle and east China can reduce the pressure of energy in the east China and the pressure of transmission and environmental protection.Furthermore, this can expedite the conversion from resource advantage to economy advantage and realize the coordinated development of nation economy. Chinese politics system,economic environment and management system also promotes UHV power grids in its development. At present, China is studying the future large power grids technology and has the ability to construct the national united power grids. On Jan, 16,2009, the first UHV power line in China was finished and put into operation.Unity or united UHV power grid, distributed power generating or scattered interactive power supplying grid are the trends of development. China, as the delegate in unity or united UHV power grid development trend, is different from any western countries. In China, is it in contradiction to develop both UHV power grid and Smart Grid? Though the large power grid with linkage effect has the advantage of optimizing the resources, it has the potential risk of power outage in large area. The ability to control the large power grid and maintain its stability is required by the fast development of power grid. And the smart power gird with self-healing and high reliability matches such requirements.Thus, smart power grid is the direction of China power grid development while building UHV power grid and the grid of different level, as well as improving the operating and management level of the grid.According to the precondition and the background of UHV power grid development in current China, the aspects which should be paid more attention on are as follows:Smart Planning: The power grid should become selfhealing and smart. The ability of power grid planning optimization should be enhanced. So should be the ability of receive-side power grid planning, on the premise of the UHV AC/ DC feed-in and differentvoltage level coordinated development. The most important thing is to change the concepts and methods of power planning, and to make the traditional power development concept such like regarding building new power stations as a wide-ranging concept of resources distribution.Smart Operation: The dispatching pattern is developing towards a coordinated control direction, aiming at the enhancement of control and mastering of large power grid. The future Smart Grid should be coordinated with a matching control center equipped with more advanced power system management ability, for the purpose of improving the functions and performance of existing EMS, MOS, WAMMAP system in an integral manner, at the same time to track down the correlations between different power grid monitoring and controlling indexes,and to construct a logic structure based power system monitoring and controlling index system. Through the gradual process of implementation of dynamic security monitoring, power system pre-alarm processing and precontrol,much more accurate and comprehensive knowledge of the operation state of the power system can be obtained, based on which, the most effective and timely measures and actions can be taken to fulfill the power system control and dispatching strategy, finally to improve the safeguard of the whole power system’s stability and security.Smart Management: The management pattern of power system is undergoing an evolution from vertical mode to distributed mode, from function management to process management, from grid construction to both construction and operation modes.V. CONCLUSIONSmart Grid is a hot spot in today's electric power system,also regarded as one of the vanes in 21st century for the major scientific and technological innovation and development in power system. Many countries in the world are involved in this big trend, and have set up a lot of Smart Grid demonstration projects and test platforms. Also, the theoretical and experimental research in Smart Grid has made some achievements. The international exchanges have greatly promoted the development of Smart Grid. Because of China's electricity distribution and extremely uneven distribution of electricity load, it is the right time to develop special highvoltage power grid. As the development of Smart Grid is still at the very beginning stage of our country, how to combine the special high-voltage power grids with intelligent power grid is the main problem confronted. The direction to development and the characteristics of smart power grids in China are still open for our experts and scholars for further study.对中国智能电网的研究摘要-智能电网是电力系统的未来发展的新方向。
第二十八章优化潮流分析(Optimal Power Flow Analysis)在对电力系统进行设计、规划和指导运行上,ETAP的优化潮流程序(OPF)是一个非常强大的模拟软件。
它在分析电力系统潮流的同时,进行系统运行条件的优化,自动调节各种控制设置,并确保符合等式和不等式的约束条件。
系统优化可以降低安装和运行成本,提高系统的整体性能,增强系统的可靠性和安全性。
除了可以减少运行和安装成本,程序还提供了各种优化目标,包括了一个整个电力系统所有的优化标准。
电力系统的任何的一个实际控制都可以通过计算来模拟,如调整变压器带载调压分接头、发电机自动电压调节控制、串并联补偿、甩负荷等。
您可以选择母线电压的约束、各种不同类型的支路潮流(视在功率MVA、有功MW、无功Mvar和电流)、以及不同的控制、约束。
OPF采用最新的优化技术,结合最先进的和最有效的计算方法,得出可靠、有效的分析结果。
经过20000母线以上大系统的测试,均能以极快的速度得出计算结果。
Operation Technology, Inc. 28-1 ETAP 7.5.0 User Guide优化潮流(OPF)程序中的主要特性总结如下:y通用和集成数据库y完全继承3维数据结构包括无限显示图,无限的联接方式,多种数据版本y处理环形,辐射型或组合系统y多个平衡节点系统y含有电岛子系统的系统y带有零阻抗支路系统y不带电母线及支路系统y根据运行温度对电缆/传输线电阻的自动调节y根据容限自动调节变压器阻抗y根据容限自动调节限流电抗器阻抗y负荷类型选择y负荷调整系数y精确的交流建模y最新的内部点处理算法y对数阻尼功能(包括等式和不等式的约束)y一次侧的双向搜寻方式y控制精度参数y系统有功功率损耗最小y系统无功功率损耗最小y平衡母线功率最小y并联无功补偿设备最少y燃料损耗最小y串联补偿最小y发电机成本最小Operation Technology, Inc. 28-2 ETAP 7.5.0 User Guidey控制运动最小y控制调整最小y电压安全指标最大y潮流安全指标最大y电压曲线平坦化y用户自定义的目标函数y母线电压约束y支路潮流约束y控制限制约束y发电机无功限制约束y数据接口限制约束y任何变量的平滑函数y发电机有功MW 和无功Mvar控制y变压器带载调压分接头控制y变压器相移控制y并联补偿控制y串联补偿控制y开关电容控制Operation Technology, Inc. 28-3 ETAP 7.5.0 User Guide28.1优化潮流工具条(Optimal Power Flow Toolbar)进入优化潮流(OPF)程序的分析模式后,优化潮流(OPF)程序的分析工具条会出现在屏幕上,这个工具条有6个功能按钮,显示如下:运行优化潮流(OPF)程序(Run Optimal Power Flow)进入优化潮流(OPF)程序的分析模式后,在工具条的分析案例选一个分析案例,然后点击运行优化潮流(OPF)按钮,进行优化潮流(OPF)的分析。
主动负荷参与配电网分布式智能电压控制徐弢;李天楚;郭凌旭;林军;张世宇;王笑雪【摘要】灵活可拓展的电压控制策略在含有多个分布式电源、且电源和负荷均不断变化的配电网具有重要意义.着眼于智能配电网的发展需求,提出一种基于案例式推理技术的分布式智能电压控制策略.通过自动感知网络电压波动,在保证系统安全和清洁能源输出比例的情况下,即时提供电压控制策略并建立完备的具有自学习能力的案例库.以包含多个分布式电源的10 kV配电网为实例,对所提出的电压控制策略进行实时数字仿真,结果证明该控制策略能有效解决网络中的电压波动问题.【期刊名称】《电力系统及其自动化学报》【年(卷),期】2016(028)001【总页数】7页(P17-23)【关键词】智能电网;分布式智能;电压控制;分布式发电;案例式推理【作者】徐弢;李天楚;郭凌旭;林军;张世宇;王笑雪【作者单位】天津大学智能电网教育部重点实验室,天津300072;海南省电力技术研究院,海口570203;国网天津市电力公司,天津300010;海南省电力技术研究院,海口570203;天津大学智能电网教育部重点实验室,天津300072;天津大学智能电网教育部重点实验室,天津300072【正文语种】中文【中图分类】TM714.2随着经济发展,负荷逐年增加以及全球节能减排战略的不断推进,分布式能源DER(distributed energy resource),特别是可再生/间歇式能源,微网MG (micro grid)以及分布式储能系统ESS(energy storage system),不断地渗透至配电网,使得配电网面临越来越多的机遇与挑战[1-4],主要表现在:DER 的并网运行使得配电网成为具有多源属性的有源配电网络,系统运行调控难度将增加。
各种随机因素和不确定因素在不同的时间和空间尺度上相互叠加,使得智能配电系统成为相互耦合的强非线性复杂系统。
新技术的应用,使得配电网可控性,可观测性和交互性有了显著提升。
Optimal Power Flow的开题报告开题报告题目:Optimal Power Flow的研究与应用一、研究背景随着电力系统的不断发展,实现可靠稳定的电力供应已经成为重要的社会需求。
由于电力系统中存在复杂的非线性特性、不确定性和耦合关系,因此研究如何减少系统的能量损耗和保证系统的可靠性是电力系统运行的重要问题之一。
为了解决这些问题,Optimal Power Flow (OPF) 作为一种重要的解决方案被提出,OPF 通过对整个电力系统进行优化调度来实现最优的电力系统运行态势。
二、研究内容本文将对 OPF 的原理、算法和应用进行深入研究。
具体研究内容包括:1. OPF 的基本理论OPF 的基本理论主要包括电力系统的基本模型、OPF 的优化目标和目标函数、该问题的优化模型以及优化算法等。
本文将着重介绍这些理论内容,阐述它们的基本思想和应用情况。
2. OPF 的算法和优化方法OPF 的算法和优化方法主要包括基本搜索算法、线性规划、非线性规划、全局优化和局部优化等。
本文将详细介绍这些方法的基本思想、原理和优缺点,以期为读者提供更多的思路和方法。
3. OPF 的应用OPF 已被广泛应用于电力系统调度、市场运营和电力市场中的竞争等方面,而且建立了成熟的理论框架。
本文将通过案例分析、实例解决方案等多种方式探讨 OPF 在以上场景中的应用情况,以及其突出的经济效益与社会价值。
三、研究目标1. 通过研究最新的 OPF 研究进展,了解该领域的基本理论和方法,并将其应用于实际问题中,提高电力系统的可靠性、安全性和经济性。
2. 拓展 OPF 的研究深度和广度,探索新的算法和模型,提高现有算法和模型的效率和精度,并为 OPF 技术的实际应用提供更加全面、细致和系统化的解决方案。
四、研究方法在研究过程中,将通过文献综述、案例分析、实例解决方案等多种方法进行探索和验证。
具体研究步骤如下:1. 搜集 OPF 的相关文献,了解该领域的研究现状等。
运筹算法分布式加速方法英文回答:Distributed Optimization Algorithms: Acceleration Methods.Distributed optimization algorithms are used to solve large-scale optimization problems that are distributed across multiple computing nodes. These algorithms can be used to solve a wide variety of problems, includingtraining machine learning models, optimizing supply chains, and scheduling tasks in a distributed system.One of the main challenges in distributed optimization is the communication overhead. Each time the nodes in the system need to share information, they must send messages over the network. This can be a significant bottleneck, especially for large-scale problems.Acceleration methods are a class of algorithms that canbe used to reduce the communication overhead in distributed optimization. These methods work by approximating the optimal solution to the optimization problem using a series of local updates. Each node in the system only needs to share information with its neighbors, and the updates can be computed in parallel.There are a number of different acceleration methods that can be used for distributed optimization. Some of the most popular methods include:Gossip averaging: This method is based on the idea of spreading information through a network by randomly exchanging messages between nodes.Conjugate gradient: This method is a classical optimization algorithm that can be used to solve a variety of problems, including distributed optimization.Nesterov's accelerated gradient: This method is a variant of the conjugate gradient method that can be used to accelerate convergence.The choice of which acceleration method to use will depend on the specific problem being solved and theavailable resources.中文回答:分布式优化算法,加速方法。
2009年第3卷第3期南方电网技术特约专稿2009,V ol. 3,No. 3 SOUTHERN POWER SYSTEM TECHNOLOGY FeaturedArticles 文章编号:1674-0629(2009)03-0007-08 中图分类号:TM934.5 文献标志码:A 法国电网电能质量承诺和电能质量评估杨显军(法国电力公司研究与开发部,92141 卡拉马特,法国)摘要:介绍法国电网的电能质量承诺,包括输电运营商(TSO)的质量承诺和配电运营商(DSO)的质量承诺。
TSO和DSO在作出电能质量承诺的同时,还利用入网规则以协议的方式来考虑和限制每个客户的负荷和分布式电源对电能质量的影响。
这些协议的实施表明,只有遵守承诺,加上电网运营商与用户双方共同作出努力,才能获得全面良好的电能质量。
对于电网干扰评估,建议使用基波潮流计算与多相谐波注入相结合的方法。
这个方法利用很短的计算时间就能对非线性负荷建模情形给出可接受的结果。
所提议的频域方法的要点在于,将每个谐波源逐一地同时作为扰动源和受扰者来考察,以便计及扰动源之间的相互作用。
研发了一个即插即用的HVDC组件并将其用在文章最后部分实例研究中。
关键词:电能质量承诺;频域建模;谐波;电力电子设备;高压直流输电Grid Power Quality Commitments and Grid Power Quality Assessmentin FranceYANG Xian-jun (Xavier X. YANG)(R&D, Electricité de France, 92141 Clamart, France)Abstract: Grid Power Quality Commitments in France are introduced, including the Power Quality Commitments of TransmissionSystem Operator (TSO) and the Power Quality Commitments of Distribution System Operator (DSO). Undertaking the commitments,TSO and DSO are using connection rules to take into account and minimize the impact on power quality of each customer (load ordistributed generation) in the way of contracts. The practice of these contracts shows that a good overall power quality can only beachieved by respecting commitments and doing effort from both of grid and customer sides. For grid disturbance assessment, it issuggested to use multi-phase harmonic injection method with fundamental power flow. This method can give an acceptable resultswith very short computing time for non-linear load modelling. The key point of proposed frequency domain method is that eachharmonic source is individually considered as both disturbance source and disturbance receiver in order to take into accountinteractions between sources. A plug & play HVDC component has been developed and used in one of two case-studies in the end ofthis article.Key words: power quality commitment; frequency domain modelling; harmonic; power electronic device; HVDC1 Power Quality Commitments1.1 Power Quality Contracts and CommitmentsStandard EN 50160 gives the main voltage pa-rameters and their permissible deviation ranges at the point of common coupling (PCC) in public low volt-age (LV) and medium voltage (MV) electricity distri-bution systems, under normal operating conditions.Based on EN 50160 and the former French power quality contract Emeraude, two new grid access con-tracts have been put into service in France between system operators (utilities) and customers: called CART for transmission customers and CARD for dis-tribution customers, see Fig. 1. These contracts not only set the power quality commitments for the grid operators but also for the customer disturbance emis-sion level.Fig. 1 Power Quality Commitments in CARD and CART8南方电网技术 第3卷The power quality that is handled in grid access contracts includes harmonic, unbalance, flicker, inter-ruption, voltage dip disturbances. Among these dis-turbances, the number of interruption is considered as commitment and others are given as indicative values. 1.2 Power Quality Commitments of Transmis-sion System Operator (TSO)French TSO has the standard and personalized contractual commitments with the customers concern-ing:1) long interruptions (1 per year, 2 per 3 years, 1 per 3 years);2) short interruptions (1 to 5 per year, 1 to 2 per 3 years);3) voltage dips (optional from 3 to 5 per year); 4) RMS voltage magnitude (± 8 %); 5) frequency (± 1 %); 6) voltage unbalance (< 2 %); 7) harmonics ;8) rapid voltage fluctuations, flicker (P lt < 1). For transmission power supply voltage, harmonic limits have been set (Tab. 1). These limits are meas-ured by 10 min mean value at each harmonic voltage (h ≤ 25, the order of harmonics) and total harmonic distortion (THD , h ≤ 40) during 100% of time. For transmission customer, current harmonic emissions are also limited in % compared to contracted rated currenth HRU lim /%h HRU lim /%h HRU lim /%5, 7 2.0 3 2.0 2 1.5 6.0 11, 13 1.5 9 1.0 4 1.0 6.0 171.015, 210.5> 40.56.019 1.0 6.0 23, 25 0.76.01.3 Power Quality Commitments of DistributionSystem Operator (DSO)The Tab. 3 shows the contracted voltage interrup-tion values for the four geographical zones.With personalized contracts signed jointly by system operators and customers, it is possible to in-clude some voltage dips in guaranteed power supply quality by network operators.Tab. 2 Transmission Customer Current Harmonic Commitments Odd Order hHRI lim /% Even Oder h HRI lim /%3 4.0 2 2 5, 7 5.04 1.0 9 2.0 > 4 0.5 11, 133.0>13 2.0Tab. 3 DSO Commitments of Voltage Interruptions per Year Interruption DurationZone 1 Zone 2 Zone 3Zone 4≥ 3 min6 3 3 2 > 1 s and < 3 min301032As grid commitments (Tab. 4), harmonic limits have been set for distribution power supply voltage. These limits are measured by 10 min mean value ateach harmonic voltage (h ≤ 25) and total harmonic distortion (THD , h ≤ 40) during 100% of time. For distribution customer, current harmonic emissions are also limited in % compared to contracted rated current (Tab. 5).Tab. 4 Distribution Voltage Harmonic Commitments Odd Harmonic OrderNot-Triplet Triplet Even HarmonicOrderTHD lim /13 3.0 8.0 17 2.0 8.0 19, 23, 251.5 8.0Tab. 5 Distribution Customer Current Harmonic Commitments Odd Order hHRI lim /% Even Oder h HRI lim /%3 4.0 2 2.0 5, 7 5.04 1.0 9 2.0 > 4 0.5 11, 133.0>13 2.0第3期杨显军:法国电网电能质量承诺和电能质量评估 9For more detail of the power qualify commit-ments of TSO and DSO, please see the following web site: /.1.4 Customer Load ConnectionIn order to meet their power quality commitments, TSO and DSO are using connection rules to take into account and minimize the impact on power quality of each customer (load or distributed generation). DSO has set up a number of technical referential documents and studying tools for load connection. Before per-forming a load or a DG connection, DSO undertakes three level’s studies which are based on standards and site measurements. The aim is to anticipate the possi-ble disturbances which may be provoked by the equipment to be connected and to determine the most adaptive Point of Common Coupling (PCC). Follow-ing these steps, DSO can ensure the optimal perform-ance of the network for all users.The first level study includes identifying the main characteristics of the installation to be connected and effectuating a global analysis. The second level con-sists of evaluating particularly the disturbances. The third level is composed of a detailed study on the iden-tified disturbances. In wind mill connection, this study is based on the standard IEC 61400-21 which evalu-ates the impact in terms of power quality [1].1.5 Power Quality Indices and Evolution ofCommitmentsIn order to measure the evolution of power qual-ity, DSO calculates a number of criterions based on network parameters. These indicators make it possible to evaluate the needs of power supply infrastructure improvement. The following values have been defined as power quality indicators:1)SAIDI or B-Criterion in France, index of mean duration of interruption for LV customers;2)SAIDI for MV customers weighted by the supplied power;3)SAIFI, indices of mean number of interrup-tion of the network, for the long interruption (t id > 3 min), the short interruption (1 s≤ t id≤ 3 min) and the very short interruption (t id < 1 s) . 2 Harmonic Source Modelling and GridDisturbance Assessment2.1 Simulation MethodFrequency domain analysis uses an admittance matrix system model developed from individual com-ponent level models connected according to system topology. The development of admittance matrix mod-els is based on multi-port network theory. Linear grid components are similarly developed from multi-port admittance parameters.In frequency domain simulation, harmonic gener-ating devices are usually modelled by current sources and the whole system admittance matrix is resolved from a start frequency to an end frequency with a fixed or variable frequency step. The deficiencies in the current source method can be partially overcome using a technique that has come to be known as “harmonic power flow”. More often, a fundamental frequency power flow solution is executed using a linear model for all power delivery equipment and loads, and the resultant fundamental frequency load terminal volt-ages are used to “adjust” non-linear load harmonic current vectors automatically without additional user action. However, in frequency approach, initial har-monic current patterns are still required to be known for each nonlinear load.2.2 Harmonic Source ModellingHarmonic model based on a constant current source, see Fig. 2(a), can’t represent real harmonic currents generated by a power electronic device con-nected to an actual network.Fig. 2 General Structure of Three Phase Harmonic Source ModelThe method (a) is too simplified because actual harmonic current of an electric device is always con-10南方电网技术 第3卷trolled by network configuration. In order to overcome this drawback, we have developed multiphase har-monic model with a voltage vector controlled current source and a frequency-variable impedance to repre-sent an actual harmonic source, see Fig. 2(b). In pro-posed method (b), J (f ) is a current source controlled by the fundamental voltage crossing frequency-variable impedance Z (f ). Z (f ) represents the behaviour of the power electronic device towards harmonic distur-bances coming from power supply side.The method (b) is detailed in the following para-graphs.Equivalent harmonic current source of a nonlin-ear load is generated by two main steps: the initial cur-rent J 0(f ) and the corrected current source J (f ) by load-flow. In parameter setting before simulation, the initial J 0(f ) is defined by pre-designed waveform pat-tern, rated power and rated voltage of the load. In fre-quency domain, J 0(f ) is represented by the followingformula:00()()cos()k kJ f J k k t ωφ=⋅+.The pre-designed waveform patterns include cur-rent waveforms generated by the most usual distur-bance sources such as 6-pulse and 12-pulse rectifiers, dimmers, arc furnaces, etc. The following procedure summarizes the proposed source modelling method:1) Choosing a pre-designed waveform (database, analytical formulas, wave form generated by Excel data sheet, empirical data table, on-site measurements) which represents the current harmonic distortion to be used;2) Refining the waveform by particular working state of the device. For example, by switching angle of rectifier and by short-circuit power of upstream net-work;3) Spectrum calculation of the refined wave-form by means of Fast Fourier Transform;4) Calculation of the designed fundamental current and phase angle between voltage and current from rated electric values of the device: voltage, pow-ers P and Q ;5) Pre-design of the harmonic spectrum at rated current and creation of a voltage controlled current source model for frequency domain simulation J 0(f ), see example in Fig. 3 for the used current pattern of a 12-pulse rectifier.Fig. 3 Current Waveform and Harmonic Spectrumof a 12-Pulse RectifierDefine the source impedance Z (f ) as a function of frequency. Z (f ) reflects the device behaviour to the disturbances coming from power supply side.For particular case, it is very important to define Z (f ) values within the studied frequencies. For instance, a static converter’s impedance was set an infinite value in the past in performing inter-harmonic studies. This is not true for a number of static converters. Some in-dustrial cases show the ripple control signal at 175 Hz can be partially absorbed by a rectifier with a capacitor filter as in a variable-speed drive [2].The proposed harmonic source model behaviours as follows: when the source J (f ) is activated, imped-ance Z (f ) is naturally set to an infinite value because the device behaviours as a current source. When the source J (f ) is not activated, Z (f ) has to be added into the system admittance matrix. In this case, the device behaviours as a passive component. Value of Z (f ) is defined case by case in accordance with studied phe-nomena:At fundamental frequency, Z (f ) is calculated from rated voltage and powers P , Q of the simulated device.第3期杨显军:法国电网电能质量承诺和电能质量评估 11At studied harmonic or inter harmonic frequen-cies whose source comes from the power supply side, Z(f) can be defined from the real response of the de-vice at these frequencies. Z(f) is depending on device structure, it can be modelled by series (or parallel) RL, RC circuits, by mixed model of CIGRE 36.05[3] or by a data table for irregular cases.If it is difficult to define a general analytical for-mula representing Z(f) for all studied frequencies. For studying a specific frequency, a simple solution is to perform a time domain simulation on the power elec-tronic device in order to obtain the impedance re-sponse. For example, at ripple control frequency of 175 Hz, the equivalent impedance at input of a recti-fier changes with the type of the rectifier structure, DC bus filter, AC filter and rectifier control techniques. A time domain simulation shows that for a 400 V, 250 kV A diode rectifier used in a variable-speed drive, the impedance at 175 Hz (Z175) is mainly determined by the DC bus type: Z175 = (0.1 ~ 0.9) × Z50 for a capaci-tive DC filter, and Z175 = (0.8 ~ 3) × Z50 for an induc-tive DC filter (Z50 is the rated fundament equivalent impedance) [2].2.3 HVDC Link Frequency Domain ModelAn High V oltage DC link (HVDC) model has been studied in frequency domain and a plug & play HVDC component has been created by using studied harmonic source models. A general HVDC link is composed of two power electronic devices (converter stations): one works as rectifier and another as inverter (or generator), see Fig. 4.Fig. 4 HVDC Link Structure between Two GridsThe studied HVDC component is decomposed in Fig. 5. Each side of HVDC is composed of a three-phase controlled harmonic current source with 74 harmonic number, an impedance and a regulator for power regulation. Two current sources are independent and each current source is synchronized with con-nected fundamental voltage after fundamental fre-quency load flow. This model can simulate independ-ently power quality and electromechanical transient phenomena of each side of HVDC link such as har-monic, reactive power, voltage fluctuation, etc. One case study below will show the application of pro-posed model.Fig. 5 Structure of Frequency Domain HVDC Model2.4 Power Quality Meter ModellingFor a power quality software, PQ meter algorithm is also a key component to think about. A general power quality software that gives just the harmonic spectrum is not enough for a site engineer to perform grid disturbance assessment in a short delay. In fact, the software needs to spend a lot of time in post-data processing in order to get necessary PQ values such as inter-harmonic, harmonic grouping, flicker, unbalance, voltage sag, etc. In most commercialised time domain software, there is no integrated power quality meter. Users have to create tool box in post data processing in order to perform grid PQ assessment. Generally, this step is time-consuming.Simulation tools for grid power quality assess-ment should include at least a build-in total harmonic distortion and individual harmonic computation unit. An advanced power quality simulation tool has to in-tegrate other disturbance metering such as in-ter-harmonic, harmonic grouping, IEC flicker meter, etc [4−5]. Tab. 6 gives some electric measurements to be integrated into power quality simulation tools.An advanced power quality meter algorithm has been studied and implanted in our power quality12 南方电网技术第3卷Tab. 6 Electric Measurements Concerning Power QualityBasic Power Quality Values Advanced Power Quality Valuestotal harmonic distortion; individual harmonic magnitude; armonic angle;voltage sag. inter-harmonic level; harmonic grouping; harmonic power; unbalance;flicker level (P st and P lt).software [6−7]. Fig. 6 shows a case study including harmonic and flicker assessments with embedded dis-turbance analyser and IEC flicker meter in a frequency domain software with electromechanical transient analysis. Harmonic distortion and flicker level Pst are displayed directly on the main simulation.Fig. 6 Disturbance Assessment with Embedded PowerQuality Meters in a Simulation Tool3 Case-Studies: Application & ValidationThe developed harmonic source models and power quality meters have been integrated in a power quality simulation tool, which has been used for grid and customer power quality analysis in a number of actual case studies. Two of them are presented here.3.1 Case Study 1: HVDC Link IntegrationA case study has been performed on harmonic level assessment and voltage improvement upon future HVDC integration to two existing grids. The study has been done using software ExpertEC [8−9] (a frequency domain software with electro-mechanical transient analysis) embedded with the developed numeric HVDC model and power quality meters. This software reads directly the substation databases and converts relative grid data into electric circuit models (Fig. 7). Current harmonic patterns at each HVDC converter station have been modelled by both simplified theo-retical data and on-site measurements offered by a manufacturer. The studied HVDC link is inserted be-tween grid 1 and grid 2 modelled by the following parameters:1)Grid 1 — short-circuit power 5 GV A, total number of nodes 1 100;2)Grid 2 — short-circuit power 1 GV A, total number of nodes 320;3)200 MW of HVDC power flow from grid 1 to grid 2;4)Frequency domain simulation: 50 ~ 3 150 Hz;5)Time step in voltage regulation 20 ms.Fig. 7 Simulated Harmonic Levels in Two GridsHVDC Option 1: Thyristor-based HVDC. Con-verter station is composed of 12-pulse rectifier + sim-plified harmonic filters with current harmonic distor-tion THD = 3.2% (Fig. 8).Fig. 8 Current Waveform and Harmonic Spectrum at Input ofThyristor-Based HVDC Converter StationHVDC Option 2: In the same grid conditions, the two converter stations are replaced by IGBT-based con-verters + harmonic filters. In this case, the current har-monic distortion TDH will be lower than 1.2% (Fig. 9).Fig. 9 Current Waveform and Harmonic Spectrum at Input ofIGBT-Based HVDC Converter Station第3期杨显军:法国电网电能质量承诺和电能质量评估 13For these two HVDC options, voltage harmonicassessment has been performed in all nodes of twogrids from 50 to 3 150 Hz. The simulation resultsshow after insertion of the HVDC option 1 (Thyris-tor-based HVDC with simplified AC filters), the totalvoltage harmonic distortion in two grids becomes toohigh (THD > 3.79%) compared to the planning level ofTHD < 3% defined by IEC 61000-3-6. The simplifiedfilters are not enough to reduce current harmonics in this case. The results from HVDC option 2 (IGBT-based HVDC) gives satisfaction: voltage har-monic levels in two grids are 0.35% for grid 1 and 0.71% for grid 2. These results mean that the integra-tion of this HVDC link will not deteriorate substan-tially harmonic levels of the two grids. This HVDC solution could be taken in consideration.Tab. 7 Total Voltage Harmonic Distortion (225 kV Level)Name of Grids Option 1, THD Option2,THD Grid 1 3.79% 0.35%Grid 2 5.91% 0.71%3.2 Case Study 2: Harmonic Current Assess-ment at PCCAnother case study has been performed on a point of common coupling (PCC) of two industrial sites. Site 1 has an important harmonic source (3 MW, power electronic devices). Site 2 has mainly linear load (induction heating by fundamental frequency) no important harmonic source but there is a very impor-tant capacitor bank for var compensation (Fig. 10).Some simulations have been done in order to confirm site engineer’s first assumption: 550 Hz har-monic source is inside site 1 although at PCC the har-monic current measured at feeder 1 is lower than that of feeder 2. Former measurements show that upstream back ground harmonic voltage at PCC is negligible if these two sites are switched off. Simulation results show that overall harmonic power flow at feeder 1 comes from downstream as the harmonic power is negative (Fig. 10). And the overall harmonic power flow at feeder 2 comes from upstream at the feeder 2 (the harmonic power is positive).Fig. 10 Simulation of Harmonic Resonance at 550 HzThis phenomenon is caused by harmonic reso-nance around 550 Hz between capacitor bank and up-stream impedance. Fig. 11 shows harmonic impedance magnitude at PCC.Fig. 11 Harmonic Impedance Magnitude at PCC4 ConclusionThis paper gives an overview of power quality commitments in grid and customer sides. Two con-tracts have been put into service in France between utilities and customers. These power quality contracts set not only the power quality commitments for the grid operator but also on the customer disturbance emission level.Generally, frequency domain software is more adequate in doing power quality assessment than time domain software. Two key points should be taken into account in power quality simulation tools: disturbance load models and power quality meter in order to re-duce computation and post processing time.Power quality meter models and three phase harmonic source models have been developed. By means of advanced build-in non-linear components, it is possible to study grid power quality assessment by14 南方电网技术第3卷dealing with non-linear loads either as disturbance sources or as disturbance receivers. As an extension of introduced general harmonic source model, an HVDC component has been built, which is composed of two independent harmonic sources and two regulators.A number of multi phase harmonic source models have been developed and integrated into a power qual-ity software, which has been used in grid harmonic analysis. Two case-studies have shown applications of harmonic source and power quality meters in actual grid power quality assessment.References:[1]NASLIN C, GONBEAU O, FRAISSE J L, et al. Wind Farm Integra-tion — Power Quality Management for French Distribution Network[C]// The 18th International Conference and Exhibition on ElectricityDistribution (CIRED 2005), June 6−9, 2005, Turin, Italy. London:IEE Conference Publications, 2005, 2: 41.[2]YANG X, DENNETIÈRE S. Modeling of the Behavior of PowerElectronic Equipment to Grid Ripple Control Signal [C]// The 7thInternational Conference on Power Systems Transients (IPST 2007),June 4−7, 2007, Lyon, France. [S.l.]: IPST, 2007.[3]O’NEILL-CARRILLO E, HEYDT G T, KOSTELICH E J, et al.Nonlinear Deterministic Modeling of Highly Varying Loads [J].IEEE Transactions on Power Delivery, 1999, 14 (2): 537−542.[4]IEC 61000-4-30:2008, Electromagnetic Compatibility (EMC): Test-ing and Measurement Techniques — Power quality measurementmethods [S].[5]IEC 61000-4-7:2002, Electromagnetic Compatibility (EMC): Testingand Measurement Techniques — General Guide on Harmonics andInterharmonics Measurements and Instrumentation, for Power Sup-ply Systems and Equipment Connected Thereto [S]. [6]IEC 61000-4-15:2003, Electromagnetic Compatibility (EMC): Test-ing and Measurement Techniques — Flickermeter — Functional andDesign Specifications [S].[7]YANG X. Methodology and Simulation Tool for Flicker PropagationAnalysis [C]// EPRI Power Quality and Applications (PQA) 2006and Advanced Distribution Automation (ADA) Joint Conference,Marriott Marquis, Atlanta, Georgia, USA, July 24−26, 2006.[8]YANG X, BERTHET L. Methodology for Frequency Domain Dis-turbance Source Modelling for Multi Source Harmonic and In-ter-harmonic Studies [C]// The 12th International Conference onHarmonics and Quality of Power (ICHQP), Cascais, Portugal, Octo-ber 1–5, 2006.[9]YANG X. Advanced Methodology and New Tool for MultiphasePower Quality Analysis & Mitigation [C]// 18th International Con-ference on Electricity Distribution (CIRED 2005), Torino, Italy, June 6−9, 2005.Received date:2009-02-28Biography:Xavier X. Yang (1958−), Male. Dr. Yang received his Electrical Engi-neering Diploma B.S. and M.S. degrees in Liaoning Technology University (China), and received the Ph.D. from National Polytechnic Institute of Tou-louse (France). Dr. Yang was employed in French Electricity and Energy Consulting Companies in which he was in charge of site measurement and power quality mitigation. He is now expert researcher in R&D of Electricite de France (EDF). His research interests are in the areas of electric power system modeling, power system harmonics, computer algorithms, fault detection and voltage sag mitigation. He is a member of IEC SC77A Work-ing Group 1 (Harmonic and flicker) and CIGRE Joint Working Groups B4 (HVDC) and C4 (Review of flicker objectives).E-mail: xavier.yang@edf.fr.收稿日期:2009-02-28作者简介:杨显军(1958–),男。
电力系统潮流计算不收敛的调整方法洪峰【摘要】潮流调整是电力系统分析计算的重要内容,潮流计算不收敛调整技术是提高分析计算自动化水平的关键.针对电网由收敛到不收敛的动态过程进行分析,提出表征电网处在不收敛临界点的综合指标,基于该指标,提出潮流不收敛调整新算法.该方法考虑了发电机启停及出力约束条件,避免了调整过程中切除负荷的情况,EPRI 36节点系统和某实际系统算例分析验证了该算法的正确性和有效性.%The power flow adjustment is an important part of the analysis and calculation of power system, and the non-convergence adjustment technology is critical to improve the automation level of the analysis and calculation. The dynamic process of power flow calculation from convergence to non-convergence was analyzed, and the comprehensive index characterized the grid at the non-convergence critical point was put forward. On the basis, a novel adjustment method was proposed in this paper. The generator start-stop and output constraints were taken into account, which avoided the adjustment process to shed load. The simulation results of the EPRI36 node system and an actual power system verified the correctness and effectiveness of the proposed algorithm.【期刊名称】《电力科学与技术学报》【年(卷),期】2017(032)003【总页数】6页(P57-62)【关键词】电力系统;潮流计算;收敛临界点;潮流调整【作者】洪峰【作者单位】湖南省电力公司建设部,湖南长沙 410004【正文语种】中文【中图分类】TM74由于在实际工作中,出现潮流无解的问题,工作人员很难区分是病态潮流还是潮流无解。
Distributed Optimal Power Flow Using ADMMTomaso ErsegheAbstract—Distributed optimal powerflow(OPF)is a challenging non-linear,non-convex problem of central importance to the future power grid.Although many approaches are currently available in the literature,these require some form of central coordination to properly work.In this paper a fully distributed and robust algo-rithm for OPF is proposed which does not require any form of central coordination.The algorithm is based upon the alternating direction multiplier method(ADMM)in a form recently proposed by the author,which,in turn,builds upon the work of Schizas et al. The approach is customized as a region-based optimization proce-dure,and it is tested in meaningful scenarios.Index Terms—Alternating direction multiplier method,decen-tralized processing,decomposition coordination,distributed pro-cessing,iterative processing,optimal powerflow.N OTATION,ADMM parameter matrices and their entries.ADMM scaled parameters.ADMM equivalence matrix.,ADMM memory matrix and parameters.Cost function at node.Aggregate cost for region.ADMM convergence measure.,Node indices.,Region indices.Number of(electrical)nodes.,Neighbors set for node and its cardinality.Lagrange multipliers for compact ADMM.,Set of regions where state is duplicated,andrelated cardinality.Projector associated with linear space.Set of non-duplicated nodes in region.Lagrange multipliers for ADMM.Active power at node.,Lower and upper limits for active power.Reactive power at node.,Lower and upper limits for reactive power. Manuscript received August03,2013;revised November18,2013and Jan-uary10,2014;accepted February11,2014.Date of publication February26, 2014;date of current version August15,2014.Paper no.TPWRS-01001-2013. The author is with the Dipartimento di Ingegneria dell’Informazione,Univer-sitádi Padova,35131Padova,Italy(e-mail:erseghe@dei.unipd.it).Digital Object Identifier10.1109/TPWRS.2014.2306495Constraint space for region.Number of regions.Region(a set).State at node.Collection of states for nodes.Constraint space for node.ADMM mixing matrix.V oltage at node.,Lower and upper limits for voltage.Wider nodes set associated with region.,ADMM global and local(region)states.,Admittance matrix and its entries.Auxiliary variable for ADMM.I.I NTRODUCTIONO NE of the key aspects in the modern convergence be-tween power electronics and communications is the pos-sibility to devise a distributed optimal powerflow(OPF)of the electrical grid by means of communication between entities and apparatuses.In this work the term distributed will denote a so-lution which is fully decentralized,which does not need any form of central coordination,which evolves by local message exchanges,and which is scalable.We assume that the commu-nication and control network is organized in(possibly overlap-ping)regions(i.e.,collection of possibly adjacent nodes),each with a local coordinator taking charge of the local processing, and of the data exchange with neighboring regions.The interest in such a challenging solution is linked to the possibility of ap-plication over large scales,and is nowadays studied in a wide variety of applicationfields,the most widely known and sim-pler formulation being the one of average consensus,i.e.,of calculating an average in a distributed fashion[4],[5].In the context of power electronics,the interest in such a distributed solution is also linked to security/privacy concerns.Privacy of information can be in fact guaranteed whenever neighboring re-gions exchange only a limited amount of information relating to the status of nodes at the boundary,which is in fact one of the targets of decentralized OPF.Distributed approaches are also naturally suited for improving system reliability in the presence of faulty processes.Over the years,a number of solutions have been proposed to tackle the distributed OPF problem.Unfortunately,the most relevant contributes(reviewed in the following),although been0885-8950©2014IEEE.Personal use is permitted,but republication/redistribution requires IEEE permission.See /publications_standards/publications/rights/index.html for more information.able to provide exceptional results in terms of accuracy and exe-cution time,are only apparently distributed.As a matter of fact, they all require some(substantial)form of central coordination. The auxiliary problem principle(APP)[6]–[10]and the al-ternating direction multiplier method(ADMM)[11]–[15],both relying on the augmented Lagrangian to improve convergence rate,were successfully applied to(OPF)in the work of Kim and Baldick[16]–[20].They both behave reasonably well with non-convex OPF problems.However,APP requires that some information is exchanged between all regions involved in the optimization process(see[18,eq.(4)]),while the classical ap-proach to ADMM adopts a scheduling of actions to solve sub-problems in a predetermined order(see[21,Algorithm2]). The alternative idea of pursuing a distributed-iterative ap-proach with incremental sub-problem solutions to speed up convergence and reduce computational load is available in the work of Conejo et al.[22]–[24],where decomposition-coor-dination is applied directly to the interior point method(IPM) used for solving the(OPF)problem.The approach employs a preconditioned conjugate gradient method to guarantee con-vergence(e.g.,see[24]),which is a form of highly centralized management.The basic algorithm of[22](without use of pre-conditioning gradient methods)is instead a fully decentralized approach which,however,fails to converge in practical net-works since[22,Theorem1,assumption(25)]hardly holds at the starting point.A similar incremental approach was recently proposed in[25],but,again,it needs solving a linear system at the central controller(see[25,eq.(19)]).An extension and optimization of Conejo’s approach to DC-OPF problems is also available in the work of Bakirtzis and Biskas[26],[27]. Some more recent efforts toward a fully distributed solution exist,e.g.,the approach discussed in[28],where authors pro-pose a decomposition of the semi definite programming(SDP) relaxation of the OPF problem.Along with the fact that SDP requires in general much more computational effort than IPM counterparts,the network decomposition is linked to maximal cliques(i.e.,not arbitrary),and the algorithm is assured to con-verge in specific contexts only(e.g.,with radial distribution net-works).In this paper we wish to investigate new approaches to the OPF problem which are completely distributed/decentralized, i.e.,do not require any form of central coordination,and are ap-plicable to any network.The solution that we propose is based upon a region-based(local)optimization process,where a lim-ited amount of information is exchanged only between neigh-boring regions in a(locally)broadcast fashion.The solution is based on ADMM in the form given by[1],which is a gener-alization of the approach of[2],[3],[29]to obtain a scalable and distributed algorithm.The result of[1]is here generalized to take into account the possibility of partitioning the problem into region-based sub-problems,which can be a more efficient solution since convergence speed may raise by reducing the number of sub-problems.The approach is robust to network par-titioning,and its convergence speed further depends upon a lim-ited number of(local)parameters which must be carefully set for an optimal behavior.To correctly run,the algorithm needs clock synchronization at the network level,that is the agreement on a common digital clock time in order to be able to concur-rently run local optimization algorithms.Such a kind of syn-chronization is a standard requirement with distributed algo-rithms(e.g.,in average consensus)and can be easily and effi-ciently achieved in a fully distributed fashion by simple mes-sage exchanges[30].The presentation of the distributed algorithm is organized as follows.Section II introduces the OPF problem and some notation about electrical quantities and region partitioning. Section III introduces the ADMM concept,which is applied to OPF in Section IV.Section V presents and discusses the resulting algorithm.Section VI discusses algorithm perfor-mance in meaningful stly,Section VII draws some conclusions.II.P ROBLEM S TATEMENTA.Standard OPF FormConsider an electrical network of nodes,. Let identify the set of(electrical)neighbors to node(in-cluding node itself),with cardinality.Let the net-work admittance matrix be,a sparse matrix which is coherent to the graph structure.Let and identify,respectively,(com-plex valued)voltage and injected current at node.The elec-trical network imposes the following powerflow constraint be-tween voltages and plex,ac-tive,and reactive powers at node are respectively denoted with ,,and.We assume a sce-nario where each node possesses both generation(G)and load (L)capabilities,and where local control on reactive poweris possible via a power electronic interface(PEI),e.g.,via the inverter used for renewable or local battery management.In this context we denote the local active and reactive generation/load range as(1) Now,the OPF problem can be formulated as a minimization problem on a cost function,which typically depends on active powers,subject to a number of constraints which include:power flow equations,active and reactive power restrictions,voltage quality restrictions.The above can be written in the form(2)A number of other constraints can also be added,e.g.,tem-poral stability constraints[31],power quality constraints[32], maximum current constraints,contingency constraints[33], etc.,all resulting in a quadratic constrained formulation.OPF istherefore a(non-convex)non-linear problem with(non-convex) quadratic constraints.B.Local FormalizationNote that(2)has a peculiar structure based on local costs and local constraints,which we want to exploit in order to obtain a formulation which is explicitly distributed.By letting the status values associated with node be the real valued couple(3) and by denoting with the collection of statuses which are neighbors to node,that is(4) we can write(2)in the form(5) where local constraints at node are collected in the (non-convex quadratic)set(6) Problem(5)fits within the local formulation of[1].C.Regional FormalizationTo obtain a more general(andflexible)result,we adapt(6) to a regional form.Define regions,.To each region associate nodes,which are(possibly overlap-ping)sets that cover the nodes set,that is,such thatIn region,define:•the state indices locally involved in the optimization process(7)•the local constraints(8)•the local aggregate cost(9)where positive constants express the fraction of cost function to be taken into account in region.For non-overlapping regions it is,since each node belongs to a unique region.When the regions are overlapping,a node can belong to more than one region,and therefore constants must satisfy andin order not to modify the aggre-gate cost,i.e.,in such a way that. Then(5)can be equivalently expressed in form(10)D.Redundant Regional FormalizationWefinally build on(10)to obtain a formalization which also fits within the ADMM setting.The idea is to associate to each region a state collecting all locally involved vari-ables,and to define the state as. We have...(11)with size2vectors representing the copy of(i.e.,of real and imaginary parts of voltage at node)available in region. Note that state entry will necessarily be duplicated in all those local states where,that is for regions belonging to set(12) whose cardinality will be henceforth denoted with. Hence problem(10)can be rewritten in the form(13) which will be our reference setting.III.O VERVIEW ON ADMMWe provide some insights on ADMM which will be used in this work(e.g.,see[15]).The standard form of ADMM solves reference problem(14) by inspecting the augmented Lagrangian(15)with an arbitrary positive constant.The search for a con-strained saddle point of is performed via an alternating pro-cedure which starts from arbitrary initial values and,and iteratively updates entries according to(16) until convergence.For a separable target function,and a sparse matrix,the ADMM iterative procedure(16)provides a distributed solution.ADMM is known to converge to a unique stable point under convexity of,,,and,and under the requirement that either is invertible or is a compact set.The limit point is assured to be a constrained saddle point for,hence the optimal solution.In a convex scenario convergence can be controlled by inspection of the measure(17) which is assured to be decreasing and converging to zero.1IV.D ESIGN OF THE D ISTRIBUTED A LGORITHMIn order to obtain a fully distributed and region-based algo-rithm,we now need to properly match ADMM,(14)–(16),to the OPF problem,(13).Formal relations are given in the following, but the reader can also skip such details andfind the resulting algorithm in Section V.pact ADMM FormulationA simpler and compact formulation for(16)is used in this work.This formulation is obtained under the assumptions that ,that is a linear space with associated orthogonal projector,and that initial choices are andfor an arbitrary value of vector.The resulting ADMM algo-rithm,which is a generalization of the approach used in[2],is proved in the Appendix.Specifically,with the above choicesis a memoryless and linear mapping of,and state can be compacted in a reduced-dimension state,to have(18) These relations allow rewriting the ADMM iterative procedure in the form(19) 1We note that the properties of(17)are related to ADMM properties discussed in[15,Proposition4.2],but they are not explicitly available in the book.For the sake of completeness,the interested reader can thereforefind a formal proof of the properties of(17)in the Appendix.with(20) where denotes a pseudo-inverse matrix operation.The starting point is an arbitrary and.Observe that the quadratic term in function acts as an attractor in the direction of,where the mixing action of on(built by message exchanges)assures coordination between different regions.The convergence measure(17)can in this context be reinter-preted by use of(18),to obtain(21) where expresses a norm weighted by matrix .For a diagonal matrix,measure can be computed at negligible cost from state and mixing values.B.Remark on ParameterA remark on the role of parameter helps in simplifying no-tation.In this respect,note that an equivalent formulation for (19),(20)is obtained by setting(22) in the definition of,and by resizing.This corresponds to neglecting the presence of.In the following we therefore adopt(22)with no loss in generality.C.Choices for Region-Based Distributed AlgorithmWe are now ready to give an explicit form to function, sets and,and matrices,,and.Function is chosen to be the separable functionthe set as the Cartesian productand to comply with the compact ADMM formula-tion.Matrix and set are chosen in such a way to force the equalities in(13),and require some more elaborate discussion. The approach which we use mimics[29]to force equality be-tween duplicate entries for each value of.The idea is to define,for each node,a symmetric weight matrix(23)where indicates the availability of a(bidirectional) communication channel from region to region to exchange information on and.We assume non-negative entries(24) and that the graph identified by is strongly connected in order to guarantee consensus for each value of,with the freedom of setting,i.e.,the absence of self-loops.Note that the required strong connectedness of is a very light request,since it is equivalent to having the information on node exchanged between neighboring regions(i.e.,those which include node in their lists).This is equivalent to a local data ex-change.Then equality in(13)is obtained by imposing(25) which forces equivalence between all entries thanks to the fact that the underlying graph is strongly connected and thatis symmetric.The peculiar form of the multiplication factor in (25)(a simpler could in principle have been chosen in place of the square root expression)is used to simplify later expressions.In order to extract the explicit expressions for matrix and projector from(25),we introduce the permutation matrix which reorders vector to group together replicas of the same quantity,that is...(26) and define matrices(27)where are column vectors.Then matrix is given in the block-diagonal form(28) with the Kronecker matrix product(to take into account the fact that state is a two-value vector).The set is chosen as the linear space(convex set)where entries assume the symmetric form in(25).By denoting with(29)the permutation matrix to swap entry with entry, the projector associated with is(30)Note that,although zero-valued entries of set some of the to zero,and although these entries are not effectively used in ADMM algorithm(19),i.e.,they could be discarded,the above(redundant)formulation allows a reasonably compact notation.The direct consequence of the above choices is a compact expression for matrix and for mixing matrix.By tedious manipulations,these take,respectively,the diagonal form(31) with the all-ones vector,and the block-diagonal form(32) Note that,for,network connectedness assures thatis positive definite,hence invertible,while this is not in general guaranteed for since choice provides.Matrices are guaranteed to have a null space associated with the all-ones vector,and for it is.V.D ISTRIBUTED OPF S OLUTIONA.Distributed AlgorithmThe proposed distributed method for OPF is obtained by ex-panding and rearranging thefindings of Section IV in a more meaningful form.The resulting algorithm is a processing par-allelized across regions,where each region concurrently runs Algorithm1,shown on the next page(see derivation details in the Appendix).Observe that,in Algorithm1,a complex-value (and compact)notation is used for voltages,that is,the real valued vector state is compacted in the complex valued scalar notation.Fur-thermore,Algorithm1is parametrized with respect to constants ,which have the property that the corresponding matrices are symmetric and identify a strongly connected graph details in Section IV-C).Con-stants identify the values and used in Algo-rithm1through the relation(33) The set(34) identifies the nodes of which have a duplicate in other re-gions.Incidentally,denote the entries of mixing vector (introduced in Section IV).Overall the algorithm requires clock synchronization to concurrently run on each re-gion.It also requires the(local)knowledge of matrices for neighboring regions.At steady state,the solution to the OPF problem(2)is pro-vided by for any,and an exit criterion can be also identified according to the standard ADMM algorithmby investigating the vanishing measure(17)[see also(21),first line]which in the present context reads as(37) Convergence can be declared when is below a given threshold which,in a per-unit system,can be set in the range depending on the desired accuracy.Measure also provides a mean of controlling the convergence process to eventually adjust parameters in order to resolve non-converging events(given by non-decreasing),the countermeasure being a simple increase of values.We observe that local problem dimensionality(in the update in line5of Algorithm1)is equivalent to the dimension of, which must be kept reasonably low for obtaining an efficient algorithm.Such a dimension is controlled whenever regions are fully(electrically)connected and have a few(electrical) links with neighboring regions(i.e.,limited cardinality of). In this case the cardinality of is close to the cardinality of the region set.This kind of choice is always possible in real net-works(since they are sparse),and further guarantees a limited amount of message exchange with neighboringnodes.The re-gion choice can be also optimized by means of sensitivity anal-ysis approaches(e.g.,see[24])to speed up convergence.Fig.1.Example of messages exchanges between regions(lines7–8of Algo-rithm1);starred values and require a two-hop communication.Afinal remark is in order with respect to the message ex-change of lines7–8in Algorithm1.We note that this corre-sponds to a light,local message exchange,as depicted in the il-lustrative example of Fig.1which shows a3-region case study. Infigure,the intra-region message exchanges are denoted with arrows.Most of them are direct exchanges from a region to a neighboring region,except made for the starred values and which denote an exchange between two non-neighboring regions(regions and in the specific case).In this latter case a simple two-hop communication is required(or,equiva-lently,a local broadcast message),this being the worst case situ-ation in the most general case.Incidentally,the local knowledge of matrices corresponds to a message exchange equivalent to that of Fig.1.B.Note on Parameter ChoiceThe selection of parameters is crucial for determining an efficient solution.Fortunately,some simple empirical choices, which become useful for improving the convergence speed,can be determined.Wefirst observe,by inspection of(35),that large values of bias the solution towards the one found at the preceding step,hence slowing convergence to the stationary point.How-ever,large values of also accelerate convergence by pre-venting the solution to be drastically different from the pre-ceding one.Because of this conflicting behavior,accurate se-lection of parameters is required.However,this rationale only applies to those boundary values which are replicated in more than one region,while for the un-replicated values(those with)a sensible choice is to set,i.e.,to remove the temporal dependency.This choice corresponds to setting(38) which is equivalent to limit the summation in(35)to. The practical consequence of this choice is that is no more positive definite(it is positive semidefinite).To assure ADMM convergence we therefore need a compact[15],which can be obtained by making a bounded set,i.e.,by simply limiting the allowable range of the entries of in(35).A second useful approach is to provide matrices,for,a meaningful but simple form.An empirical setting that hasshown to provide good convergence speed is(39) where is the Kronecker delta providing1for and0 otherwise.Global constant is selected for good convergence speed,and is a weight associated with region. Weights need to be chosen according to the amount of power generated in the region,in such a way that a larger generated power corresponds to a smaller associated weight,that is,a smaller dependence in function.Recalling that indicates the availability of a communication channel from region to region,choice(39)corresponds to a(locally) fully connected graph associated with.Incidentally,choice(39)satisfies(38).C.On Algorithm ConvergenceSince ADMM is assured to converge only in convex sce-narios,and OPF is a non-convex problem,the distributed solu-tion is in general not guaranteed to converge.However,a con-vergence guarantee can be given under some(weak)requests on the solvability of sub-problems(35).Specifically,the following result holds,whose validity is proved in the Appendix. Theorem1:Assume that sub-problem(35)has zero primal-dual gap when written in the equivalent form(40) with variable duplicated.We then have:1)if original problem(13)has zero primal-dual gap,then Al-gorithm1is assured to converge to a primal-dual solution, and defined in(37)is assured to be a decreasing func-tion converging to0;2)if original problem(13)has non-zero primal-dual gap,thenAlgorithm1diverges in either or,or both,anddiverges,too.Note that the zero primal-dual gap request on(40)is equiv-alent to asking that the problem is solvable via standard IPMs, which are commonly adopted for OPF problems,and are known to be sufficiently robust to OPF non-convexities,as well as to variable duplication.As a consequence,the ADMM algorithm exhibits the same limitations of centralized IPM counterparts, and is therefore expected to behave well in practice.From our simulation experience,the request on(40)is met most of the times,and global convergence is also assured.VI.S IMULATION R ESULTSA.Simulation SettingsWe test the proposed distributed method with the wide area networks of the IEEE Power System Test Case Archive[34]. Network sizes of,57,and118are used.Active power constraints are selected according to available generation power as defined in test archive,thatis,.TABLE IR EGIONS C HOICE IN IEEE N ETWORKSThe limits imposed to reactive power at generators are,[90,90]kV Ar,and[110, 110]kV Ar,respectively,for,57,and118.Load power is set according to the test archive,and linear costs are assumed,unless otherwise stated.In order to set active voltage constraints,a normalized voltage rangeis assumed for and118,while for the range is relaxed(widen)to[0.9,1.1]to assure the existence of a feasible solution.Matrices are given in the form(38)and(39).The chosen region partitioning and associated weight choices are reported in Table I,with regions for,and regions for and 118.Regions are non-overlapping.Note that lower weights have been assigned to regions associated with a higher presence of PEIs(e.g.,generators,static V AR compensators, etc.).Parameter is selected for good convergence speed,to have,135and65,respectively,for,57,and118. In order to test algorithm robustness,the starting point is set to the(non-optimal)voltage value given in the test archive,which is an unfeasible starting point far from the optimal solution. Overall,the three tests correspond to a simpler scenario with moderate region partitioning,a more elaborate sce-nario with moderate region partitioning,high number of nodes, and presence of many loops,and a challenging scenario with severe region partitioning,uneven distribution of generation resources,and complex network dependencies .In all three cases we are dealing with a non-convex OPF problem.B.Performance Evaluation ExamplesFig.2shows the algorithm behavior with above settings. Specifically,in thefirst column of Fig.2we display the value of the overall generation power as a function of the iteration number,and the derived using a standard centralized solver is reported in dash-dotted lines(we used IPOPT[35]).Observe that in all cases the optimum valueFig.2.Performance of distributed OPF with IEEE networks:(a)-(d)30-bus system with4regions and;(e)-(h)57-bus system with8regions and;(i)-(l)118-bus system with4regions and.Performance measures shown infigure are:(a),(e),(i)overall generation power;(b),(f),(j)local reactive power generation;(c),(g),(k)limit voltage value of network nodes;(d),(h),(l)mean squared difference measure as defined in(37).is reached.Power generation is a good measure for controllingalgorithm convergence,but any other electrical parameterwould display similarly,as can be observed in the second column of Fig.2reporting the reactive power generation associated with PEIs as a function of the iteration number. The limit value for network voltages are instead reported in the third column of Fig.2,from which we can observe the correct action of voltage limitations.In the fourth column of Fig.2wefinally show the mean squared difference as defined in(37).exhibits an almost piecewise linear convergence rate.The overall rate is slower with due to the higher partitioning(8regions instead of4).We also observe that varies along the time scale,with a much higher initial speed,and with possible speed changes along the way(the variation being more evident for). This effect is known in ADMM and is due to the specific local structure of costs and constraints.Note that the decreasing be-havior of in Fig.2assures the convergence of the distributed process.Note also that small glitches in the decreasing behavior of(more evident for)are due to non-convexities of the OPF problem,and to local instances where the assumptions of Theorem1do not hold.parison With a Centralized SolutionTable IIfurther provides some comparison with a standard centralized solution.The centralized algorithm of[24]was selected,which is an IPM-like approach using the unlimited point algorithm(UPA).TABLE IIC OMPARISON W ITH C ENTRALIZED S OLUTION[24]A preconditioned conjugate gradient method(GMRES)is em-ployed,with the further help of inertia correction and of a suit-able choice of the initial point(according to[35]).All such highly centralized tools are necessary to assure convergence of the algorithm.On the distributed ADMM side,the algorithm is implemented in MatLab,and a MatLab interface to IPOPT[35] is used to solve sub problems(35).MUMPS linear solver is used within IPOPT.The target level for declaring convergence is set to with the further requirement that constraints must be satisfied within a precision of in order to assure that a feasible point is obtained.In Table II we observe that the high accuracy reached( with respect to the value obtained with the centralized al-gorithm)is far below any practical requirement,and certifies the validity of the distributed method.A comparison in terms of computing time is also available.Note that the processing time required by taking advantage of parallel processing(i.e.,。