【作者】 李杨；

【导师】 顾雪平；

【作者基本信息】 华北电力大学 ， 电力系统及其自动化， 2014， 博士

【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。更多还原

【Abstract】 Transient stability assessment (TSA) has been being an important task to ensure the secure and economical operation of power systems. With interconnection of large-scale power grids, electricity market reform and growing presence of large-scale intermittent renewable energy, the dynamic behaviors of the power systems are becoming more complex and difficult to be controlled, with more serious consequences resulted from loss of stability. The existing TSA methods, such as time domain simulation methods and direct methods, can not meet the needs of online applications required by the modern power systems. In recent years, pattern recognition-based TSA (PRTSA) technology has attracted great attention of researchers in such a field at home and abroad. Its main task is to establish a mapping relationship between system state variables and system stability conclusions. Compared to the other TSA methods, the PRTSA methods have a lot of advantages, such as strong learning ability, fast assessment speed and acquisition of potentially useful information. They have a good prospect in the field of the on-line security and stability analysis of power systems.This thesis systematically studies the relative problems of PRTSA, including feature selection, classifier construction, online learning, topology change adaptation and rule extraction. First, an original feature set is extracted from the post-fault system information which can be provided by the Wide-Area Measurement Systems (WAMS) to represent the characteristics of power system transient stability, and an optimal feature subset is then selected by a proposed feature selection method to reduce the input space dimension. Then, a TSA classifier based on the optimized extreme learning machine (ELM) is constructed. After that, an online learning mechanism based on the integrated online sequential ELM is proposed. Finally, the adaptability of the proposed TSA method to the network topology changes and extraction of assessment rules from the ELM-based classifier are studied.The main work of this thesis is summarized as follows:1. A new feature selection method based on an improved maximal relevance and minimal redundancy (mRMR) criterion is proposed for TSA. First, based on the post-fault system information provided possibly by WAMS, an original feature set for stability classification is extracted. Then, the standard mRMR is improved and applied to feature selection for compressing the feature set. A group of nested candidate feature subsets are obtained by using the incremental search technique, and each candidate feature subset is evaluated by a support vector machine classifier to find the optimal feature subset with the highest classification accuracy.2. A novel TSA model based on an optimized ELM is proposed. Based on the selected optimal feature subset, ELM is used to build a TSA classifier. The parameters of the ELM model are optimized by the improved bacterial colony chemotaxis algorithm, and the classification ability of the ELM model is improved.3. A TSA online learning mechanism based on an ensemble of online sequential ELM (OS-ELM) model is proposed. In order to overcome the difficulty of on-line model updating, the OS-ELM and online boosting algorithm are employed as a weak classifier and an ensemble learning algorithm for online learning of the integrated ELM models respectively. The stability and generalization ability of the OS-ELM model is greatly improved.4. The adaptability of the proposed TSA method to the network topology changes is investigated. Based on the original feature set established, a sample set is constructed with the network topology changes in consideration, and the optimal feature subset is found by using the proposed feature selection method. The optimized ELM is then employed as a TSA classifier to evaluate the network topology adaptability of the proposed TSA method. The results show that, compared to the previous TSA methods, the topology adaptability of the proposed method is much higher.5. A novel rule extraction method for TSA is proposed by using ELM and an improved Ant-miner algorithm. In order to overcome the shortcomings of low understandability and interpretability of a’black-box’learning machine, first the basic principle of Ant-miner algorithm is studied; then based on the selected optimal feature subset, an example sample set is generated by the trained ELM-based TSA model; and finally, a set of classification rules are obtained by the improved Ant-miner algorithm to replace the original ELM network for TSA.更多还原