遗传算法在暂态稳定评估输入特征选择中的应用

针对主成分分析中利用传统方法进行特征选择的缺陷,提出了基于遗传算法的特征选择方法。选择反映电力系统运行状态的特征变量,建立暂态稳定评估模型;为了提高数据处理的效率,首先对原始数据进行了动态聚类分析;对数据进行主成分分析后,以类内类间距离判据作为适应度函数,采用二进制编码形式的遗传算法进行特征选择。通过对3机9节点和10机39节点新英格兰系统的计算,验证了所选方法的有效性。

Feature selection based on genetic algorithm for transient stability assessment

Aimed at the disadvantages existing in feature selection by traditional combination optimization method in PCA(Principal Component Analysis), a new method based on genetic algorithm to select the input features is put forward. In this approach, the feature set to describe the system status and post-fault network configuration change are selected for transient stability assessment and the initial data is preprocessed by dynamic clustering analysis firstly. With the within-class/between-class distance criterion used as fitness function, a binary genetic algorithm is employed to select an effective subset of features forming the feature set after PCA, and the input dimension is reduced remarkably. As an example, the 3-machine 9-bus WSCC system and the 10-machine 39-bus New England system are used for simulation. The results reveals the validity of the proposed approach.