摘要
为进一步提升电力系统暂态稳定评估的准确率,依据电力系统暂态过程数据的时序特性,建立了一种基于双向长短期记忆(Bi-LSTM)网络的暂态稳定评估模型。该方法通过Bi-LSTM网络建立底层量测数据与电力系统暂态稳定类别之间的非线性映射关系,采用准确率、F1指标和FPR指标综合评估Bi-LSTM网络模型性能的优劣,在此基础上,采用t分布随机近邻嵌入(t-SNE)降维方法和k最近邻(KNN)分类器进一步提升暂态稳定评估的准确率。新英格兰10机39节点系统算例表明:所提模型比传统的机器学习模型和部分深度学习模型拥有更好的评估性能。通过可视化方法和网络预测分数对评估模型进行分析,结果表明Bi-LSTM网络模型具有较强的电力系统暂态过程特征提取能力,适用于电力系统暂态稳定性的评估。进一步研究了底层输入数据的归一化模式和方法对暂态评估模型的影响,结果表明z-score归一化方法要优于min-max归一化方法,采用总维数归一化模式的模型评估性能更好。
In order to further improve the accuracy of transient stability assessment(TSA),a TSA model based on the bidirectional long-short-term memory(Bi-LSTM)network is established according to the sequential characteristics of data in the power system transient process.This method uses the Bi-LSTM network to establish a non-linear mapping relationship between the basic measurement data and the transient stability category of power system.The performance of Bi-LSTM network model is evaluated by accuracy,F1-measure(F1)index and false positive rate(FPR).On this basis,the t-distribution stochastic neighbor embedding(t-SNE)dimension-reduction method and the k-nearest neighbor(KNN)classifier are used to further improve the accuracy of TSA.The example based on the New England 10-generator 39-bus system show that the proposed method has better performance than conventional machine learning models and some deep learning models.The assessment model is analyzed by visualization methods and network prediction scores.The results show that the Bi-LSTM network has a strong ability to extract the characteristics of power system transient process,which is suitable for the TSA of power system.Further,the influence of the normalization mode and method of basic input data on the TSA model is studied.The results show that the z-score normalization method is better than min-max normalization method,and the assessment performance of the model using the total dimension normalization mode is better.
作者
孙黎霞
白景涛
周照宇
赵晨昀
SUN Lixia;BAI Jingtao;ZHOU Zhaoyu;ZHAO Chenyun(College of Energy and Electrical Engineering,Hohai University,Nanjing 211100,China)
出处
《电力系统自动化》
EI
CSCD
北大核心
2020年第13期64-72,共9页
Automation of Electric Power Systems
基金
国家自然科学基金资助项目(51707056)。
关键词
深度学习
长短期记忆网络
暂态稳定评估
归一化
t分布随机近邻嵌入
k最近邻
deep learning
long-short-term memory(LSTM)network
transient stability assessment
normalization
t-distribution stochastic neighbor embedding(t-SNE)
k-nearest neighbor(KNN)
