改进Sammon映射算法在分析暂态稳定评估输入特征有效性中的应用

基于机器学习技术的电力系统暂态稳定评估方法中,输入特征提取的是否合理往往决定了最终的分类效果。然而,目前却缺乏一种工具去评价选择的输入特征是否具有可分性。鉴于此,引入Sammon映射算法将高维样本数据映射到低维空间中,通过观察映射点的分布情况判断提取的特征是否有效,并针对原算法的不足之处进行改进。首先利用主成分分析法(principal component analysis,PCA)求出包含原始数据信息最多的前两维主成分向量,代替原算法随机取值的方法,作为映射点坐标向量的初始值。然后,采用迭代修正法求解最终的映射点坐标向量,加快了求解速度。最后,以改进Sammon映射算法作为工具,分析IEEE 39节点系统的仿真数据和某地区实际在线历史数据提取特征的有效性,证明该算法在指导特征选择中具有良好的应用前景。

Application of Improved Sammon Mapping Algorithm in Input Features Validity Analysis of Transient Stability Assessment

In the method of power system transient stability assessment based on machine learning technology, the reasonableness of the input feature extraction decides the final classification result. However, there were no tools to judge whether the selected input features are the separable. Therefore, this paper introduces the Sammon mapping algorithm to map high dimensional sample data to low dimensional space, determines the effectiveness of selected feature through observing the distribution of mapping points, and improves the original algorithm according to its deficiencies. Firstly, we adopted principal component analysis ( PCA) method to obtain the first two dimensional principal component vectors containing the most original data information, which worked as the initial value of the mapping point coordinate vector instead of the random selection method in the original algorithm. Then, we used the iterative method to solve the coordinate vector of mapping points to accelerate the solving speed. Finally, we used the improved Sammon mapping algorithm as a tool to analyze the effectiveness of selected features of the numerical simulation data in IEEE39-bus system and the actual online historical data of a certain area. The analysis results show that the improved algorithm has a good application prospect in guiding feature selection.