自组织映射神经网络用于暂态稳定性分析的研究

对几种形式的自组织映射神经网络进行了集中介绍,并对自组织特征映射（SOFM）神经网络和学习矢量量化（LVQ）神经网络在电力系统暂态稳定模式识别中的应用性能进行比较。利用SOFM网络输出层聚类信息对不同ANN输入特征量的选取效果进行了直观的比较。在这些比较的基础上,利用Kohonen网络“无监督聚类、有监督学习”的工作方式,给出一种基于Kohonen网络的复杂系统在线事故筛选和发电机功角预测方法。利用华中电网的数据对这种网络进行了大量的计算,计算证实了该方法的有效性。

SELF-ORGANIZING MAPPING (SOM) NEURAL NETWORKS FOR POWER SYSTEM TRANSIENT STABILITY ASSESSMENT

In this paper, several self-organizing mapping (SOM) neural networks are introduced. They are the common Kohonen network, its modified model which works in an "unsupervised grouping, supervised learning" way, the common LVQ (learning vector quantization) network, and the network model that combines Kohonen and LVQ. Their performance on transient stability assessment is compared and the second model is chosen in later study for its high reliability. This paper also compares the different ANN input attributes' abilities to represent the patterns in the real world, using the visual clustering maps on the output layers of self-organizing feature mapping (SOFM) neural networks. Based on these analyses, a novel online large power systems' contingency screening and rotor angle predicting approach is proposed and tested. All the simulation data comes from Central China Power Grid and the simulation results verify the effectiveness of the method.