基于模态分析和Relief算法的在线静态电压稳定特征选取方法

Feature Selection Method for Online Static Voltage Stability Based on Modal Analysis and Relief Algorithm

输入特征变量选择和降维是基于决策树静态电压稳定评估面临的重要问题之一,电压稳定评估结果的效率和准确率很大程度上取决于所选特征变量集合的合理性,该文提出了一种能有效降低维数的特征变量选择方法。该方法首先依靠模态分析完成初步的筛选,提取出导致电压失稳的关键影响因素;其次根据Relief特征选择算法进一步优化,给每个特征变量一个权重值,排除权重值较低的特征变量。最后以实际电网为案例,利用决策树十折交叉法对评估结果进行验证。结果表明,以该方法对静态电压稳定性进行特征变量选择,得到的最简组合在预测静态电压稳定性上具有较高的准确率和较短的建模时间。

The selection and dimensionality-reduction of input features is one of the key issues related to the assessment on static voltage stability based on decision tree. Considering that the efficiency and accuracy of assessment results largely depends on the rationality of selected feature set, a feature selection method which can effectively reduce the number of dimensionalities is proposed. Firstly, a preliminary screening is done by using modal analysis to extract the key factors leading to vohage instability. Secondly ,the extracted factors are further optimized by using Relief feature selection algorithm, thus a weight value is given for each feature variable, and those with lower weights are excluded. Fi- nally, a decision tree based lO-fold cross validation method is used to verify the assessment results of an actual power network. The results show that by using the feature selection method for static voltage stability, the accuracy of the obtained simplest combination in the prediction of static voltage stability is higher, and the modeling time is shorter.