【作者】 黄彦全；

【导师】 肖建；

【作者基本信息】 西南交通大学 ， 电力系统及其自动化， 2005， 博士

【摘要】 电力系统状态估计是电力系统调度、控制、安全评估等方面的基础,也是电能管理系统的核心组成部分。自状态估计方法引进到电力系统领域后,在电力系统状态估计涉及的各个方面,如估计准则、不良数据检测和辨识、不良数据修正、系统可观测性和量测配置、状态估计计算的稳定性、带约束条件的状态估计方法、状态估计的分块和并行计算方法、配电网络的状态估计方法、电力系统状态抗差估计方法、加权估计方法中权值的选择和权函数的研究、系统参数的估计和辨识、动态电力系统状态估计方法以及各种新技术和新理论的应用等,取得了丰硕的研究成果,大大提高了电力系统状态估计的技术水平。同时,我们也应该看到,电力系统状态估计的各个方面依然存在着许多尚未解决的问题,随着技术水平的提高和对电力系统状态估计要求的变化,电力系统状态估计在某些方面也需要进行一定的改进。 论文的研究工作围绕着电力系统状态估计准则和迭代计算、大规模电力网络分析和状态估计、不良数据检测和辨识、状态运动轨迹的回归和预测等方面展开。论文首先综述了电力系统状态估计的一般方法,就等效电流量测变换状态估计、电力系统带约束状态估计、正交变换在电力系统状态估计的应用进行了讨论。 大规模电力系统情况下,网络的分块和等值是提高电力系统分析效率和实时性的有效途径。在详细分析研究已有等值和分块算法的基础上,提出了基于支路切割网络分块的电力系统分析算法,使大规模电力系统分析的问题可以通过若干较小规模系统的分析和协调变量的计算获得解决,该算法具有简单、与原有潮流计算方法兼容、易于实现分布式计算和提高对大规模系统分析计算速度的特点,仿真试验结果表明了算法的有效性。 提出了一种基于支路切割网络分块的电力系统状态估计新算法。在基于支路切割网络分块潮流分析计算方法的基础上,通过支路切割,把电力系统划分为若干较小规模的子系统,通过异步方式交替对子网络进行状态估计迭代。讨论了算法所涉及到的参考节点问题、不良数据检测和辨识问题和解决方案。 不良数据检测和辨识是电力系统状态估计的重要内容,为此提出了一种不良数据检测和辨识新算法。现有的估计计算后进行不良数据检测和辨识算法难以避免“残差淹没”和“残差转移”的现象,从而难以一次性地获得最更多还原

【Abstract】 Power system state estimation is the core of electric energy management system and the bases of dispatch, control, security evaluation and so on. Since state estimation was introduced into power systems, many results have been obtained in aspects related to power system state estimation. The harvests can be witnessed in estimation criterion, detection and identification of bad data, correction of bad data, observability of system, measurement configuration as while as in stability of state estimation algorithms, state estimation with equality and inequality constraints, state estimation blocking and parallel calculation, state estimation of distribution networks, robust state estimation of power system. Researches on the determination of weight value and weight function in weighted estimation algorithm, estimation and identification of system parameters, together with dynamic power system state estimation and various subjects on the applications of new technique and theory could also be the markers of the improvements.However, problems still exist in certain aspects of power system state estimation. As the improvement of techniques and the changes of requirements on power system state estimation, amelioration becomes necessary in some fields of power system.This thesis focus on the problems, including the principle of state estimation, iteration calculation, state estimation and analysis in large scale power network, detection and identification of bad data, regression and prediction of state trajectories. Ordinary method of power system state estimation is first discussed in this thesis, and then discussion was carried out on equivalent current measurement transformation of power system state estimation, constrained power system state estimation and the applications of orthogonal transformation in power system state estimation.In large scale power systems, network blocking and equivalence analysis is an effective way to enhance real-time applications and the efficiency of power system analysis. On the basis of previous equivalence and blocking algorithms, a network blocking algorithm of power system analysis based on branch cutting is proposed, which would make large scale power system analysis get solved byanalyzing some small scale power system and calculating the consistent variables. This algorithm is simple, and can be compatible with previous power flow calculation method, what is more, it can be easily to realize distributed computing and accelerate the computation speed of large scale power system analysis. Simulation results show the validity of this algorithm.A new algorithm of power system state estimation based on network blocking using branch cutting method is suggested. On the basis of power flow calculation and network blocking using branch cutting method, the power system is divided into several smaller subsystems, so that state estimation iterations can be processed asynchronously. Problems related to this algorithm about reference bus, detection and identification of bad data, together with the resolutions are discussed.Detection and identification of bad data is a key factor in power system state estimation. A new algorithm of detection and identification of bad data is detailed. The current used algorithm, which detects and identifies bad data after estimation, is difficult to avoid residual mask and residual transfer phenomenon, so it is hard to obtain optimal estimation results. Through analyzing the correlativity of measurement and the correlativity of residual, a method to detect bad data is presented, making use of correlation coefficient of measurement variations. Simulation results show that this algorithm is able to distinguish bad data from sudden change data, meanwhile it can identify multi bad data once with low failing ratio. According to the correlativity of the measurements, topology structure and parameters of system, a new method is brought forward to identify measurement noise in the collection of suspicious bad data by means of estimation. The two algorithms constitute an integrated way to detect, identify and correct bad data before state estimation, which effectively guarantees the validity and efficiency of the state estimation.Applications of various new theories and techniques are one of the ways to accelerate the development of power system state estimation. Support vector machine regression is one of the popular regressions based on statistical-learning theory. Algorithms based on support vector machine regression in power system state estimation are introduced. Support vector machine regression and least square support vector machine regression are respectively used to accomplish one-step prediction of the system state, and then the iteration calculation of system statescan be achieved utilizing previous state estimation algorithms. Considering the operation characteristic of power system, support vector machine regression and least square support vector machine regression are applied to fulfill the training of active power model and reactive power model, to accelerate model training. According to the prediction of system state using the trained model, bad data will be detected and identified.Simulation results indicate that support vector machine regression is of superior performances in state tracking, noise-tolerant and robustness.更多还原