【作者】 华栋；

【导师】 陈皓勇；

【作者基本信息】 华南理工大学 ， 电力系统及其自动化， 2012， 博士

【摘要】 我国电力市场化改革已经迈出了实质性的步伐，但仍处于起步阶段，电力市场的交易规则、监管制度、市场机制等问题有待进一步研究和完善。电力市场环境下各市场主体的复杂关联和交互作用可采用博弈论来进行建模，而实际电力市场往往过于复杂，无法直接采用标准的博弈论模型来进行分析。特别是市场主体反复观察市场并且根据市场情况不断调整自身策略的能力增加了问题的复杂性，使常规方法更加难以用来进行电力市场分析。基于代理的计算经济学和实验经济学扩展了传统的博弈论分析方法，为电力市场提供了全新的分析手段。同时论文综合分析国内外电力市场成功与失败的经验，认为在目前条件下建立电力市场对电网公司来说有一定的风险性。因此，为了将风险化解在方案设计和规则制定阶段，在电力市场正式启动之前，有必要密切结合电网实际建立一套电力市场综合模拟系统，用于研究、验证各种市场方案和规则的合理性及分析可能出现的风险，并在市场启动运营前为各市场成员进行培训提供一种有效的技术支持手段。为此本文对电力市场仿真进行了深入研究，且旨在设计并开发一套电力市场综合模拟系统。在系统构建过程中，本文对以下关键问题提出新颖的解决方法，主要工作如下：1.人工智能中强化学习的RE学习算法、使代理能够通过不断地与环境进行交互所得到的经验中进行学习，适合在电力市场智能模拟中运用。在电力市场综合模拟系统中实现了发电商代理基于RE学习算法。实验结果表明运用基于RE学习算法可以较好地模拟发电商的经济特性。2.基于代理的计算经济学已经成为电力市场研究的一种重要方法。但作为一种新的方法，还有许多问题没有解决，其中一个基本的问题是研究方法的问题，包括仿真实验的设计方法，仿真结果的分析方法等。以电力市场研究中常用的一种代理学习算法—RE学习算法为例，基于一个拥有四个发电商的日前电力市场，进行了市场模型、智能代理模型、实验目的及实验方案的设计。设计了三组实验，前两个实验根据实验结果对基于代理的电力市场仿真实验轮数的设定、实验盘数的设定等问题进行了分析和讨论，后一个实验对RE算法中的一些参数的选取问题进行了探讨，展示了其在选取时的矛盾性，并从加强学习法的机理上分析了其原因。3.大规模风电的发展使得现行的电力市场设计，规则都要发生改变，不仅要给风能发电及传统发电形式同样的激励，也要保证尽可能的降低社会成本，让终端用户的电费合理。提出了几种大规模风电接入系统的电力市场设计模式，并对这几种模式进行了比较研究；应用混合整数规划法求解大规模风电并网时电力市场的机组组合问题，证明了当风电功率预测偏差较大时候，会使市场交易成本增加。4.研究不同负荷—价格响应特性电力市场的市场均衡问题，以及对不同的负荷—价格响应特性电力市场,分别比较统一价格拍卖和歧视性价格拍卖对市场均衡结果的影响。实验结果说明了电力市场中引入可中断负荷和需求侧参与竞争的重要性，在设计市场制度时要充分考虑不同需求弹性的影响。5.完成了电力市场仿真系统的整体结构设计和功能模块设计，该系统具备配置各种实验方案、修改实验参数、记录并处理实验结果数据等功能，可以提供基于实验经济学、ACE和基于人工—代理混合的三种仿真方式，在国内外电力市场仿真系统设计中尚属首例。提出了完整的电力市场仿真实验设计方法及评价仿真实验结果的市场评价指标；并就测试系统的实验结果进行分析与评价，并将系统用于模拟实际的西北电力市场。更多还原

【Abstract】 China has launched a marketization reform of power industry, but, it’s still at thebeginning stage. Therefore, it’s necessary to further study and revise the market mechanismand regulatory system of electricity markets. The complicated correlations and interactions ofmarket participants under electricity markets can be modeled with game theory, unfortunately,the practical electricity markets are always too complex to conveniently modeled usingstandard game theoretic techniques. Especially the market participants observe the marketrepeatedly and adjust their strategies continuously according to the market situations, whichadds the complexity of the problem and makes the normal methods difficult to analyzeelectricity markets. Agent-based computational economics (ACE) and experimentaleconomics extend the traditional game theoretic approaches and provide new methods forelectricity market analysis.After analyzing the success and failure of market operation experience home and abroad,it’s really a risk for the grid companies to construct power market at present. Consequently, inorder to avoid the risk at the stage of market mechanism designing and regulation making, it’snecessary to design and develop an integrated electricity market simulation system combinedwith the practical power grid. The simulation system also can be applied in studying andverifying the rationality of market mechanisms and regulations, analyzing the probable riskand supplying a technique support system for the market participants training before theformal operation of electricity market.In order to design and develop a electricity market integrated simulation system, thesimulation approaches of power market are studied deeply, some key issues are solved in anovel way, the main research work includes:1. RE learning algorithm which belong to reinforcement learning algorithm, are adaptiveand can find the optimal strategy of the dynamic system through experience obtained from thedirect interaction with its environment, the algorithm also has the quality of low requirementfor prior knowledge. The feature makes them well suitable for dealing with thedecision-making problems of power suppliers. Program module which help the agents makedecision based on RE learning algorithm, has been realized in the Electricity Market Integrated Simulation System. The experimental results show that these agent learningalgorithm are capable of making the agent simulate the economic characteristics of powersuppliers and can reflect the different bidding psychology of the power suppliers.2. With the development of large-scale wind power, the market design and ruls willchange, which gives the same incentives to wind power and traditional power suppliers. Themarket design should lower the social cost and set the end-user electricity fees reasonable.The article proposed several power market design modes with large-scale wind powerparticipating, and made the comparsion among these modes; the article also adopted MIPmethod to solve the unit commitment problem with large-scale wind power connected to thesystem, and proved the increasing of deviation of wind power forcasting would meanwhileincrease the trading cost in market.3. The article studied the market equilibrium in different load-price reactioncharacteristics in power market, and compared the effect on market equilibrium underuniform price auction and discriminatory price auction modes respectively. The resultsrepresented the importance of the participating of interrupted load an demand side in powermarket.4. The integral architecture design and function modules design of Electricity MarketIntegrated Simulation System are described. The simulation system has such functions asfollows: setting kinds of experiment schemes and experiment parameters; recording andanalyzing experiment results; supply three kinds of simulation methods which includesimulation based on human bidding, agent-based simulation and combination of agent-basedand human method, which is proposed for the first time in the market simulation systemresearch home and abroad. Then simulation experiment organizing and experiment resultsanalyzing methods are proposed, and the simulation system is applied to analysis a testingsystem market and a certain practical regional power market.更多还原