高压交流输电线路故障特征挖掘与故障原因辨识

发布时间：2018-03-14 02:51

本文选题：输电线路故障　切入点：故障原因辨识　出处：《山东大学》2017年硕士论文　论文类型：学位论文

【摘要】：输电线路是电网的重要组成部分,覆盖范围广,运行环境恶劣复杂,极易由于自然灾害、人为破坏等原因而发生故障。作为省级电网的骨干电力网络,220kV及以上的高压交流输电线路的故障必然给电网带来冲击,威胁电网的安全稳定运行。同时,继电保护装置的快速动作可能导致故障线路损坏痕迹不明显,增加了故障检修困难程度。在线路跳闸后,准确、及时地判断出故障的可能诱因同时结合测距位置以及当地地形状况,可指导巡线,快速、准确地发现故障点,减少故障排除时间,提高电力系统的供电可靠性及和运行稳定性,具有重大的经济效益及社会效益。目前国内外的辨识研究较少,多为针对某种故障的规律统计以及防治措施建立,尚无系统的辨识方法。本文针对输电线路常见的雷击、风偏、鸟闪、污闪、树闪以及山火六种原因引起的输电线路单相故障进行研究。在各故障类型故障机理分析的基础上,对故障特征规律进行深入剖析,结合历史经验对故障相关的外部特征以及内部特征进行总结。外部特征包括故障发生时对应的天气、季节以及时间特征,内部特征则指反映故障本质的故障重合闸情况、故障相电流直流含量、三次谐波含量、过零点畸变情况以及过渡电阻大小和性质,并通过实际故障录波数据的处理分析进行验证。更进一步,本文采用基于Fisher分数的特征挖掘方法实现对六种故障模型辨识而言各特征的重要性程度计算并进行排序,从而探究出各故障原因类型辨识的主要影响因素,有针对性地建立不同的故障原因辨识模型。鉴于输电线路故障样本的不完备,本文采用具有强泛化能力的支持向量机(SVM)实现故障原因的分类辨识,算法以结构风险最小化为目标,在一定程度上避免了过学习问题。针对六种故障原因分别带入相关特征,利用样本进行训练建立辨识模型,并且使用粒子群算法(PSO)对各模型参数进行优化。预测阶段,将待测故障样本进行相应的特征分析,并分别带入该六种故障模型,得出对应于各种故障类型的概率,取其中的最大值所对应的故障原因类型为判别结果并进行验证测试。并对于算法提出了基于样本数据不平衡问题以及算法自学习能力的改进。最后通过测试结果表明,基于PSO优化的SVM算法辨识效果得到了提升,并且通过选取特征重要性排序靠前的特征可以在保证辨识准确率的同时简化模型计算量,提高效率。综上,本文基于雷击、风偏、鸟闪、污闪、树闪以及山火六大输电线路故障的综合分析研究,提出一种综合故障外部特征以及故障内部特征的PSO-SVM故障原因自学习辨识的方法。本方法建立在对实际故障数据的挖掘分析的基础上,理论依据充分,算法仿真证明准确率高。同时本方法所用故障数据信息容易获取,应用时可结合当地故障规律进行扩展,能够实现对常见故障原因的有效识别,满足工程实际要求。
[Abstract]:Transmission line is an important part of the power network. It covers a wide area and runs in a harsh and complex environment, so it is easy to be caused by natural disasters. The failure of HVAC transmission line, which is the backbone of the provincial power network, will inevitably impact the power network and threaten the safe and stable operation of the power network. The rapid action of relay protection device may lead to the failure line damage trace is not obvious, increase the trouble degree of fault maintenance. After the line tripping, accurate, The possible cause of fault can be judged in time by combining location of location and local terrain, which can guide the inspection line, find fault point quickly and accurately, reduce the time of troubleshooting, improve the reliability of power supply and operation stability of power system. It has great economic and social benefits. At present, there are few researches on identification at home and abroad, most of them are statistics and prevention measures for certain faults, and there is no systematic identification method. In this paper, the common lightning strike and wind deviation of transmission lines are discussed. The single-phase fault of transmission line caused by bird flicker, pollution flashover, tree flash and hill fire is studied. Based on the analysis of fault mechanism of each fault type, the fault characteristic law is deeply analyzed. Combined with historical experience, the external and internal characteristics of the fault are summarized. The external features include the weather, season and time characteristics corresponding to the occurrence of the fault, while the internal features refer to the fault reclosing situation, which reflects the nature of the fault. The DC content of the fault phase current, the third harmonic content, the distortion after 00:00 and the size and properties of the transition resistance are verified by the processing and analysis of the actual fault recording data. In this paper, Fisher score based feature mining method is used to calculate and sort the importance of each feature to identify six fault models, so as to find out the main factors affecting the identification of each fault cause type. In view of the incomplete fault samples of transmission lines, support vector machine (SVM), which has strong generalization ability, is used to classify and identify the fault causes. The algorithm aims at minimizing structural risk and avoids the problem of overlearning to a certain extent. PSO is used to optimize the parameters of each model. In the stage of prediction, the fault samples under test are analyzed and brought into the six fault models respectively, and the probability corresponding to various fault types is obtained. Taking the fault cause type corresponding to the maximum value of the algorithm as the discriminant result and the verification test, the problem of unbalance based on the sample data and the improvement of the self-learning ability of the algorithm are proposed. Finally, the test results show that, The identification effect of SVM algorithm based on PSO optimization is improved, and by selecting the feature of feature importance ranking, we can simplify the calculation of the model and improve the efficiency while ensuring the accuracy of identification. Comprehensive analysis and research on the faults of six transmission lines, bird flash, pollution flashover, tree flash and mountain fire, In this paper, a method of self-learning identification of PSO-SVM fault cause based on external and internal fault features is proposed. The method is based on the mining and analysis of actual fault data, and the theoretical basis is sufficient. At the same time, the fault data information used in this method is easy to obtain, and can be extended in combination with the local fault law, which can effectively identify the common fault causes and meet the practical engineering requirements.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TM755

【相似文献】