基于数据挖掘的静态电压稳定在线评估

发布时间：2018-11-07 07:05

【摘要】：随着社会经济的发展和环境因素的制约,电力系统运行越来越接近稳定极限,高渗透率可再生能源的大规模接入,增加了电力系统运行的复杂性和不确定性,也对电力系统电压稳定评估提出了新的要求。传统的静态电压稳定评估方法因计算耗时、建模困难,难以满足在线评估的应用要求。随着向量测量单元(phasor measurement unit,PMU)的广泛使用,海量的电网实时数据为电压稳定的在线分析提供了可能性。本文提出一种基于数据挖掘的静态电压稳定在线监测算法,核心思想是通过离线仿真分析产生大量的原始数据,并应用机器学习的方式,从大量数据中提取出有价值的信息,再通过PMU实现在线监控的目的。首先利用电力系统静态电压稳定的分析方法,通过PV曲线计算不同网架下基于电压稳定约束的最大传输功率和电压稳定储备系数。并在PV曲线极限点处求出各母线节点对主导电压失稳模式的参与因子,以及各母线的电压-无功灵敏度。基于此,引入模糊聚类分析的方法对电压稳定指标进行综合判断,从而更准确地识别电压薄弱区域,并应用中国某地区电网数据进行算例分析及验证。在对整个电网的静态电压稳定评估的基础之上,展开数据挖掘。针对较多的输入特征变量使得模型训练时间长以及分类准确率低的问题,本文从电压失稳的实质出发,在充分考虑导致电压失稳原因的基础上来选择输入变量。首先依靠模态分析确定电力系统电压稳定问题的主要影响因素,完成初步筛选,其次根据Relief特征选择算法进一步优化,最终得到最佳的特征变量集合,从而降低电力系统的特征维度。最后选择决策树模型作为静态电压稳定评估的分类器,并引入代价敏感机制,提出了基于代价敏感决策树的静态电压稳定在线评估算法。该算法以误分代价最小为目标,可在一定程度上避免电压失稳漏诊为电压稳定的情形,有效降低了漏警率。电网调度人员可利用PMU实时采集需要监测的变量数据,依据决策树中提取的判定规则,对电力系统的静态电压稳定性进行快速评估,实现在线监测的目的,并应用某地区电网数据进行算例分析及验证。
[Abstract]:With the development of social economy and the restriction of environmental factors, the operation of power system is more and more close to the limit of stability. The large-scale access of renewable energy with high permeability increases the complexity and uncertainty of power system operation. New requirements for voltage stability evaluation of power system are also put forward. The traditional static voltage stability evaluation method is difficult to meet the requirements of online evaluation because of its time-consuming calculation and difficult modeling. With the wide use of vector measurement unit (phasor measurement unit,PMU), massive real-time data of power grid provide the possibility of on-line analysis of voltage stability. In this paper, a static voltage stability on-line monitoring algorithm based on data mining is proposed. The core idea is to generate a large amount of raw data by off-line simulation analysis, and to extract valuable information from a large number of data by means of machine learning. Then the purpose of online monitoring is realized by PMU. Firstly, using the static voltage stability analysis method of power system, the maximum transmission power and voltage stability reserve coefficient based on voltage stability constraints under different grid structures are calculated by PV curve. At the limit point of the PV curve, the participation factors of each bus node to the dominant voltage instability mode and the voltage-reactive power sensitivity of each bus are obtained. Based on this, the fuzzy cluster analysis method is introduced to judge the voltage stability index synthetically, so as to identify the weak voltage area more accurately, and to use the data of a certain area in China to analyze and verify the example. Based on the static voltage stability evaluation of the whole power network, data mining is carried out. In view of the problem that more input characteristic variables make the training time of the model long and the classification accuracy low, this paper selects input variables based on the essence of voltage instability and considering the causes of voltage instability. Firstly, the main influencing factors of voltage stability in power system are determined by modal analysis, and the primary screening is completed. Secondly, the optimal set of characteristic variables is obtained by further optimization according to the Relief feature selection algorithm. Thus, the characteristic dimension of power system is reduced. Finally, the decision tree model is selected as the classifier of static voltage stability evaluation, and the cost sensitive mechanism is introduced, and an on-line static voltage stability evaluation algorithm based on the cost sensitive decision tree is proposed. The algorithm aims at minimizing the cost of misdivision, which can avoid the situation that voltage instability is diagnosed as voltage stability to a certain extent, and effectively reduce the leakage alarm rate. The power grid dispatcher can use PMU to collect the variable data that needs to be monitored in real time, according to the decision rules extracted from the decision tree, the static voltage stability of power system can be evaluated quickly, and the purpose of on-line monitoring can be realized. An example is used to analyze and verify the data of a regional power network.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM712;TP311.13

【参考文献】