含风电电力系统的场景分析方法及其在随机优化中的应用

发布时间：2018-01-10 02:04

本文关键词：含风电电力系统的场景分析方法及其在随机优化中的应用　出处：《武汉大学》2014年博士论文　论文类型：学位论文

【摘要】：风力发电具有显著的随机性和波动性,且不受调度。随着我国风电并网比例与日俱增,风电“并网难”问题日益突出。在电力系统规划与运行中如何充分考虑风电的随机性和波动性,成为了当前世界范围内工业界和学术界普遍关心的前沿性难题。论文首先构建了基于场景分析方法的含大规模风电电力系统随机优化决策框架,根据是否考虑随机变量的相关性,将风电场景细化为静态场景和动态场景两方面分别论述场景生成的方法,并结合各自的具体实例——大规模风电输送通道落点优选和随机机组组合,深入研究了场景分析方法在随机优化中的应用,取得了以下几方面的创新成果：在场景生成方面：风功率预测误差的解析理论分布对于不同的预测手段和应用地点尚不具有广泛的适用性。对此,本文提出了“以随机变量的经验分布作为场景生成的输入”的思想；通过等间距抽样法,对单个随机变量或多个相互独立的随机变量的经验分布进行抽样,采用柯列斯基分解或场景树法对抽样值进行重新排列组合,介绍了一套基于拉丁超立方抽样的静态场景生成方法；已有的动态场景生成方法没有计及风电的波动特性。并且,与目前广泛使用的风功率“点预测”手段相应的动态场景生成方法不完备。对此,本文提出了一种考虑风功率随机性和波动性的动态场景生成方法,采用“预测箱”统计风功率点预测的预测误差经验分布,通过对多元正态分布协方差结构的关键参数进行辨识和逆变换抽样,使得随机生成的动态场景既符合风电的随机性又符合波动性。关于风功率静态场景在随机优化中的应用方面：论文将静态场景生成与随机优化方法结合应用到我国某省接入外来大规模风电的输送通道落点优选问题中,设计了风电落点多目标决策的评价指标体系及其权重设置方法；通过仿真分析,本文发现电力系统的多样化运行方式可能会影响决策模型的结果。因此,论文利用带有概率信息的多个典型静态场景刻画了运行方式的多样性,提出了一种考虑多场景的风电落点随机优化方法。关于风功率动态场景在随机优化中的应用：论文以随机机组组合作为对象,通过计算发现：经过经典的场景削减方法得到的最有可能发生的场景可能忽略部分极端场景,这些极端场景虽然发生概率很低,但是一旦发生造成的影响是巨大的。鉴于此,本文从场景生成得到的原始大量场景集合中辨识出极端边界场景,并将其引入到修正后的随机机组组合模型中刻画极端事件造成的损失期望,构成了两阶段随机机组组合方法,权衡了系统运行的经济性和可靠性。并且随机机组组合模型经过混合整数线性化,调用CPLEX软件实现了问题的快速准确求解。
[Abstract]:Wind power generation has significant randomness and volatility, and is not scheduled. With the increasing proportion of wind power grid in China. The problem of "grid-connected difficulty" of wind power is becoming increasingly prominent. How to fully consider the randomness and volatility of wind power in power system planning and operation. Has become the world-wide industry and academia generally concerned about the forefront of the problem. Firstly, a stochastic optimization decision framework for wind power system with large-scale wind power system is constructed based on scenario analysis method, according to whether or not to consider the correlation of random variables. The wind power scene is divided into static scene and dynamic scene, and the methods of scene generation are discussed respectively, and combined with their specific examples, large-scale wind power transmission channel drop point selection and random unit combination. The application of scene analysis method in random optimization is deeply studied, and some innovative results are obtained as follows: In the aspect of scene generation, the analytic theoretical distribution of wind power prediction error is not widely applicable to different prediction methods and application sites. In this paper, the idea of "taking the empirical distribution of random variables as the input of scene generation" is put forward. The empirical distribution of a single random variable or several independent random variables is sampled by the method of equidistant sampling, and the sampling values are rearranged and combined by Kiresky decomposition or scene tree method. A static scene generation method based on Latin hypercube sampling is introduced. The existing dynamic scene generation methods do not take into account the fluctuating characteristics of wind power. Moreover, the dynamic scene generation methods corresponding to the wind power "point prediction" which are widely used at present are not complete. In this paper, a dynamic scene generation method considering the randomness and fluctuation of wind power is proposed. The prediction error empirical distribution of wind power point prediction is calculated by "forecasting box". By identifying and sampling the key parameters of multivariate normal distribution covariance structure, the randomly generated dynamic scene is not only consistent with the randomness of wind power, but also in line with the volatility of wind power. On the application of static wind power scene in random optimization: this paper combines static scene generation with stochastic optimization method to select the location of transmission channel in a province of our country. The evaluation index system and its weight setting method of multi-objective decision making of wind power drop point are designed. Through simulation analysis, it is found that the diversified operation mode of power system may affect the result of decision model. Therefore, the paper describes the diversity of operation mode by using several typical static scenes with probability information. In this paper, a stochastic optimization method for wind power drop points considering multiple scenarios is proposed. About the application of wind power dynamic scene in stochastic optimization: this paper takes the random unit as the object. It is found by calculation that the most likely scenarios obtained by the classical scenario reduction method may ignore some extreme scenarios, although the probability of these extreme scenarios is very low. In view of this, the extreme boundary scene is identified from the original set of a large number of scenarios generated by the scene. It is introduced to the modified stochastic unit commitment model to describe the loss expectation caused by extreme events, which constitutes a two-stage stochastic unit combination method. The economy and reliability of the system are weighed, and the stochastic unit model is linearized by mixed integers, and the CPLEX software is used to solve the problem quickly and accurately.
【学位授予单位】：武汉大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM614

【参考文献】