基于双向长短期记忆网络的输电线路状态画像与评估

Transmission line status portrait and assessment based on bidirectional long shorttime memory networks

为提高输电线路状态评估的准确率,提出一种先聚类再回归的输电线路状态画像与评估模型。首先,设计自组织神经网络对输电线路原始数据进行降维,自适应地提取若干类代表性特征信息,无需人工提取特征和依据主观经验选择聚类数;然后,将代表性特征数据输入LSTM(长短期记忆)网络中,LSTM网络将前向学习和反向学习相结合,对模型进行双向训练与评估,建立输电线路核心数据与状态的非线性映射关系,提高电网场景下的输电线路状态评估准确率。实验结果表明,所提模型在实际数据集上取得了较好的评估效果,评估准确率高于常用的支持向量机、人工神经网络、稀疏自动编码机等方法。

To improve the transmission line status evaluation accuracy,the paper proposes a transmission line status portrait and assessment model based on clustering and later regression. Firstly,the self-organizing neural network(SONN)is designed to reduce the dimensionality of the original data of the transmission lines and to adaptively extract several types of representative feature information without manual feature extraction and selection of the number of clusters based on subjective experience. Secondly,the representative data is fed into the LSTM(long shortterm memory)networks. The networks combine forward learning and reverse learning to conduct bidirectional training,evaluate the model,establish the nonlinear mapping relationship between the core data and the transmission line state,and improve the evaluation accuracy of the state of the transmission line in the power grid scenario. The experimental results show that the model proposed in this paper has achieved good evaluation results on the actual data set;specifically,it is superior to conventional support vector machine,artificial neural network,sparse automatic encoding machine,and other methods in evaluation accuracy.