论文部分内容阅读
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。
In the power system transient stability assessment method based on machine learning technology, whether the input feature extraction is reasonable or not usually determines the final classification result. However, at present, there is a lack of tools to evaluate whether the selected input features are divisible. In view of this, the Sammon mapping algorithm is introduced to map the high-dimensional sample data into the low-dimensional space. The distribution of the mapping points is observed to judge whether the extracted features are valid or not, and the defects of the original algorithm are improved. First, the principal component analysis (PCA) is used to find the first two-dimensional principal component vectors that contain the most original data information, instead of using the original algorithm randomly, as the initial value of the mapping point coordinate vector. Then, using iterative correction method to solve the final mapping point coordinate vector, speed up the solution. Finally, the improved Sammon mapping algorithm is used as a tool to analyze the simulation data of the IEEE 39-bus system and the validity of the extracted features of the actual online historical data in a certain area. The results show that the proposed algorithm has a good application prospect in guiding the feature selection.