基于卷积Transformer的支路参数辨识
Branch parameter identification based on convolution Transformer
王琳鹏 1宋公飞 1汪梦龙1
作者信息
- 1. 南京信息工程大学自动化学院,南京 210044
- 折叠
摘要
为了解决在电力系统支路参数辨识任务中传统的单支路特征参数辨识只能对单一目标进行识别,无法充分利用电力系统的历史信息的问题.在Transformer模型基础之上采用一种卷积自注意力,帮助模型更好地捕获输入特征之间的关系,将局部上下文更好地纳入注意力机制;同时采用卷积门控循环单元进行位置编码,保证内容和位置的一致性,从而降低训练损失,进一步改善预测的精确度.仿真结果表明,该算法预测精度较高,明显优于其它的机器学习算法和深度学习算法.
Abstract
In order to solve the problem that the traditional single-branch feature parameter identification can only identify a single target,but cannot make full use of the historical information of the power system.Based on the Transformer model,a convolution self-attention is used to help the model better capture the re-lationship between input features and better incorporate the local context into the attention mechanism.At the same time,the convolution gating loop unit is used for position coding to ensure the consistency of con-tent and position,thus reducing the training loss and further improving the prediction accuracy.The simula-tion results show that the prediction accuracy of this algorithm is higher than other machine learning algo-rithms and deep learning algorithms.
关键词
深度学习/卷积自注意力/相对位置编码/参数辨识/输电线路Key words
deep learning/convolution self-attention/relative position coding strategy/parameter identifi-cation/transmission lines引用本文复制引用
基金项目
国家自然科学基金资助项目(61973170)
国家自然科学基金资助项目(61973168)
出版年
2024
NETL
NSTL
万方数据