【作者】 殷桂梁；

【导师】 史锦珊；

【作者基本信息】 燕山大学 ， 测试计量技术及仪器， 2006， 博士

【摘要】 随着分布式发电在电力生产中所占比重的增加，对电力系统的影响越来越大，分布式发电与大电网的并列运行将引起一系列的问题，其中最重要的问题之一即为分布式发电系统的孤岛检测。孤岛检测方法主要可分为两大类——有源孤岛检测和无源孤岛检测。分布式发电系统有源孤岛检测方法虽然灵敏度高，但不可避免地将引起系统的供电质量下降和系统的不稳定，而现有的无源孤岛检测方法都存在非检测区大的缺点。本文在现有文献的基础上，针对分布式发电系统中谐波模式随分布式电源和负荷不断变化的问题，提出了基于人工免疫系统谐波功率模式识别的分布式发电系统无源孤岛检测方法。首先研究分布式发电系统中谐波模式的表示方法，分析了电流谐波功率的计算和谐波功率状态空间映射。为减小状态空间的维数，将分布式发电系统中奇次谐波频谱分段积分，映射到二维状态空间。然后采用模式识别中的聚类分析研究了分布式发电系统的孤岛检测方法。阴性选择算法是免疫系统识别外来抗原最基本的机理，本文通过分析人工免疫系统中阴性选择算法的二进制匹配规则，比较r-连续匹配规则、r-块匹配规则和Hamming距离匹配规则的性能，提出了具有检测规则的阴性选择算法，并采用遗传算法生成检测规则，从而使生成的规则数量最少并达到很好的非我状态空间覆盖。详细分析了生物免疫系统B细胞和T细胞识别外来抗原的机理，研究了B细胞基因重组和超变异机制，T细胞识别抗原时与B细胞的相互作用，提出了实现孤岛检测的T模块检测器和B模块检测器模型。T检测器直接用于检测分布式发电系统的孤岛谐波功率模式，T模块可以采用随机初始化方式生成，通过在孤岛检测过程不断学习新的谐波模式和调整其运行工作点，适应分布式发电系统中谐波模式的变化。B模块不直接用于检测孤岛运行状态的谐波模式，而是通过变异和克隆机制，与测量得到的外部编码向量相互作用，将其结果送到T模块，从而改进T模块在状态空间中的覆盖。为控制B模块的数量，B模块的克隆向量引入了死亡机制，这样实现了用有限的检测器检测几乎无限的电流谐波功率模式，并将算法的复杂程度限制到可实际应用的大小。实际应用中，要得到充分的分布式发电系统孤岛运行时的谐波模式一般很困难，因此本文将分布式发电系统并网运行的谐波模式作为有导师学习算法的输入，采用实数阴性选择算法提取孤岛谐波模式的非我样本。运用数值估计的方法，根据分布式发电系统并网运行的自我样本子集，估计出自我状态空间体积，计算出抗体覆盖非我状态空间的数量。然后采用模拟退火算法优化抗体集合在非我状态空间中的分布，通过神经网络分类算法实现孤岛模式的检测。本文采用电力系统电磁暂态仿真软件PSCAD和Matlab对所提出的孤岛检测算法进行了仿真计算，仿真结果证明了算法的正确性。更多还原

【Abstract】 As the contribution of distributed generation to the electric power production increases, the effects on the power system grow more important. There will be a series of issues when the distributed generations are grid-connected. The one of the most important is the islanding detection. There are two kinds of islanding detection method: active islanding detection and passive islanding detection. Although the active islanding detection methods have high sensitive, the power quality degrading and system unstability will have to be added to the system by them. And there are non-detection zones in all current passive islanding methods. The distributed generation passive islanding detection methods based on the harmonic power pattern recognizing using artificial immune system are presented in the paper for solving issues of the harmonic pattern changes as the distributed generators and loads turn on or off.The harmonic patterns in distributed generation system are studied firstly. The algorithm of the current harmonic powers and the mapping of the state space of the harmonic power are analysized.The mapping of the harmonic powers state space using the integral of odd harmonic frequency spectrums is presented. An islanding detection method using clustering analysis is studied.The negative selecton is the basic mechanism to recognize intrusive foreign antigens for the nature immune system. The binary matching rules of the negative selection are analysized in this paper. The performance of the r-contiguous matching, r-chunk matching and Hamming distance maching rules is compared. A negative selection algorithm with detection rules is presented. Using genetic algorithm to evolve rules to cover the non-self space, the number of rules is very small as well as a good covering of the non-self space is gotten.According to the mechanisms of B celll and T cell recognizing the foreign antigen, the gene rearrangement and somatic mutation of B cell, T cell and B cell interaction as T cell recognizing antigen, the T module detectors and B module detectors are presented. The T-module is initialized by choosing points at random and can learn new patterns and tune the detection position when it is used to detect islanding. Using mutation and clone mechanism the B-module reacts to all frequently occurring state vector values and presents its results to the T-module. The B-module also plays a role in updating the T-module. A death mechanism is also introduced for the clone vectors of B-module for controlling the population of B-modules. Thereafter, the algorithms of islanding detection can almostly detection all the current harmonic power pattern. And the complexity of the algorithm is accessible for practical application.In practical application, it is difficult to get the enough pattern samples of distributed generaton islanding state. An approach to get the islanding harmonic pattern using the real negative selection algorithm is presented. The input of the algorithm is the harmonic patterns of the grid-connected distributed generation system. The volume of the self state space is caculated using the numerical analysis. Then, the distribution of antibody sets in the self sated space is optimized using simulated annealing algorithm. The islanding patterns is detected using artificial neural nentwork classifier.By using PSCAD and Matlab, various islanding operation conditions and normal load variation are simulated. And all the algorithms proposed in this paper are verified.更多还原