基于广域动态信息的电力系统暂态稳定评估研究

发布时间：2018-01-25 00:04

本文关键词： 暂态稳定评估广域测量系统特征选择极限学习机在线学习规则提取　出处：《华北电力大学》2014年博士论文　论文类型：学位论文

【摘要】：电力系统暂态稳定评估(TSA)一直是关系到电力系统安全稳定运行的重要问题。随着大区电网互联、电力市场化改革和大规模可再生能源的接入,系统的动态行为更加复杂多变,控制变得更加困难,电网暂态稳定破坏的后果也更加严重。时域仿真法、直接法等现有的TSA方法,难以满足电网运行对在线稳定评估的要求。近年来,基于模式识别技术的TSA方法(PRTSA)受到各国学者的广泛关注,取得了较大的进展。其主要任务是建立系统变量和系统稳定结果问的关系映射,具有学习能力强、评估速度快、能提供潜在有用信息等优势,在电网在线安全稳定分析领域有着良好的应用前景。本文系统地研究了PRTSA的特征选择、分类器构建、在线学习、拓扑变化适应性及规则提取等问题。首先,研究从广域测量系统(WAMS)可提供的故障后信息中抽取有效表征系统暂态稳定性的模式特征集,通过特征选择方法筛选出最优特征子集,降低输入空间维数；然后研究暂态稳定评估分类器的构建,提出一种基于优化极限学习机的暂态稳定评估模型；接着研究评估模型的在线学习机制,提出一种基于集成在线序贯极限学习机的暂态稳定评估方法；最后研究了暂态稳定评估网络拓扑变化的适应性及暂态稳定规则的提取问题。论文的主要研究成果包括： 1、提出一种基于改进最大相关最小冗余(mRMR)判据的TSA特征选择方法。首先,基于WAMS可提供的故障后信息,建立稳定分类的原始特征集,然后对mRMR判据进行改进后应用于特征选择和特征集压缩。通过增量搜索算法得到一组嵌套的候选特征子集,并使用支持向量机分类器验证各候选特征子集的分类性能,选择得到具有最大分类正确率的最优特征子集。 2、提出一种基于优化极限学习机(ELM)的暂态稳定评估模型。基于所选的最优特征子集,采用极限学习机来构建TSA分类器,并采用基于综合混沌搜索策略的改进细菌群体趋药性算法优化选取ELM模型的参数,提升了评估模型的分类能力。 3、提出一种基于集成在线序贯ELM的评估模型在线学习机制。针对评估模型不能在线更新的不足,采用增量式学习的在线序贯ELM作为弱分类器、在线Boosting算法作为集成学习算法进行多ELM模型的在线集成学习,提高了在线序贯ELM的稳定性和泛化能力。 4、对暂态稳定评估方法的拓扑变化适应性进行了研究。基于本文依据WAMS信息建立的原始特征集,构造考虑网络拓扑变化的样本集,并采用本文的特征选择方法得到考虑系统拓扑变化的最优特征子集,然后采用ELM构建TSA模型对本文提出方法的拓扑变化适应性进行了研究评价。结果表明与已有PRTSA方法相比,本文方法适应电网拓扑变化的能力具有显著改进。 5、提出一种基于极限学习机和改进蚁群挖掘算法的稳定评估规则提取方法。为了克服“黑箱型”学习机可理解性差、解释性差的缺陷,首先研究了蚁群挖掘算法进行规则挖掘的基本原理；然后基于所选最优特征子集,从训练好的ELM中产生示例样本集；最后,采用改进蚁群挖掘算法从示例样本集产生一组可以替代原ELM网络的分类规则。
[Abstract]:Power system transient stability assessment (TSA) has been an important issue for the security and stabilization of power system. With the relation to the regional power grid interconnection, access power market reform and large-scale renewable energy, the dynamic behavior of the system is more complicated, the control becomes more difficult, transient stability and failure consequences are more serious. The time domain simulation the TSA method, the existing method of direct method, it is difficult to meet the power grid operation evaluation of on-line stability. In recent years, the TSA method based on pattern recognition technology (PRTSA) has attracted wide attention of scholars, has made great progress. Its main task is to establish the mapping relationship between system variables and system stability results has asked, strong learning ability, evaluation speed, can provide useful information and other advantages, in the power grid online stability analysis has a good application prospect.
This paper systematically studies the PRTSA feature selection, classifier construction, online learning, topology adaptability and rule extraction. Firstly, from the research of wide area measurement system (WAMS) can provide the transient stability after fault information extraction system model effectively characterize the feature set, the best subset of features selected by feature selection method. To reduce the dimension of input space; and then study the transient stability classifier evaluation, put forward an evaluation model of transient stability optimization based on extreme learning machine; then study the evaluation model of online learning mechanism, a method is presented to assess the transient stability of integrated online sequential extreme learning machine based on the extraction of; finally the transient stability assessment and transient adaptability stable rules of network topology. The main contributions of this paper include:
1, put forward an improved optimization based on TSA feature selection method (mRMR) criterion. Firstly, WAMS can provide the fault information based on the original feature, establish a stable classification set, and then the mRMR criterion is applied to feature selection and feature set compression. Improved search algorithm to get a set of candidate feature subsets nested by incremental, and using a support vector machine classifier to verify each candidate feature subset selection, get the feature with the highest classification accuracy subset.
2, put forward a kind of learning machine based on Optimization limit (ELM) for transient stability assessment model. The optimal feature subset selection based on TSA classifier to construct the extreme learning machine, and using the improved bacterial colony chemotaxis integrated chaotic searching algorithm is used to select ELM resistance parameter model based on improved classification ability assessment model.
3, put forward an evaluation model of ELM integrated online sequential mechanism based on online learning. Aiming at the shortage of evaluation model can not be updated online, the online sequential ELM incremental learning as a weak classifier, online Boosting algorithm as the ensemble learning algorithm for multi ELM model integrated learning in line, improve the stability and generalization ability of online sequential ELM.
4, changes in topology adaptability to transient stability assessment methods were studied. The original features based on the WAMS information set is established based on the network topology changes the sample set by the consideration of the structure, and get the optimal feature selection method considering the system topology changes sign subsets, and then build on the proposed topology change adaptation method the study and evaluation of TSA model by ELM. The results showed that compared with the existing PRTSA method, this method has the ability to adapt to changes in network topology significantly improved.
5, put forward a method of extracting stability evaluation rules of extreme learning machine and improved ant colony algorithm based on mining. In order to overcome the "black box" learning machine understandable, explain the defects of the poor, the basic principle of the first ant colony mining algorithm of rule mining; then based on the feature subset is generated sample sample set from trained ELM; finally, the improved ant colony algorithm for mining classification rules generated a set of alternative ELM network from the sample set.

【学位授予单位】：华北电力大学
【学位级别】：博士
【学位授予年份】：2014
【分类号】：TM712

【参考文献】