基于转子角轨迹簇特征的电力系统暂态稳定评估

Power System Transient Stability Assessment Based on Cluster Features of Rotor Angle Trajectories

机器学习技术已被广泛应用于暂态稳定分析领域。在基于机器学习的暂稳评估中,如何兼顾输入特征信息量的多少和整体计算效率,一直是需要解决的问题。为此提出一种基于转子角轨迹簇特征、由线性支持向量机(linear support vector machine,LSVM)和决策树(decision tree,DT)构成的组合式暂稳评估方法。首先,构建转子角轨迹簇整体特征的时间序列作为暂稳评估的输入向量,考虑到输入特征的时间维度,先通过LSVM对每个时序特征进行降维映射,再将降维后的结果输入至DT中,形成暂稳预测和稳定程度评估模型,并采用boosting技术进一步提高评估模型的准确性。对新英格兰10机39节点系统进行算例分析验证了方法的有效性,所提出的轨迹簇特征和组合算法具有较高的精度和计算效率,能较准确地指示系统的稳定程度,且对未知运行工况具有一定的泛化能力。

Multiple machine learning techniques have been widely used in transient stability analysis.For machine learning based method,balance between input feature number and total calculation efficiency is always a problem need to solve.In this paper,a hybrid classifier combining linear support vector machine（LSVM） and decision tree（DT）was proposed to assess transient stability using rotor angle trajectory cluster features.Firstly,rotor angle cluster features were used as inputs.Considering time dimension of input features,each time series feature was reduced with LSVM.Then the reduced data were put into DT to generate transient stability prediction and stability degree evaluation models.Boosting technique was used to improve accuracy of the evaluation model.Case studies were conducted on New England 10-machine 39-bus system to verify the proposed method.Test results showed that the proposed cluster features and algorithm possesses high accuracy and overall calculation efficiency.The evaluation model could indicate stability degree accurately and was robust to untrained samples.