基于注意力机制和双向长短期记忆网络的电能质量扰动识别

Recognition of Disturbances in Power Quality Signals Based on Attention Mechanism and Bi-directional Long Short-Term Memory

提出一种基于深度学习的电能质量扰动信号分类新方法.该方法应用注意力机制和双向长短期记忆网络构建分类模型,并利用Matlab仿真产生训练数据集与验证数据集.考虑了7种常见的复合扰动信号,并将其作为序列数据直接输入到网络中进行训练和验证.实验结果表明,本方法能准确地识别不同的扰动信号,在验证集上模型的识别准确率可以达到99.7%.通过对比实验发现,应用注意力机制和双向长短期记忆网络的模型的识别能力要优于支持向量机和多层感知机等传统机器学习算法.

In this study,basing on deep learning we propose a new model that can perfectly classify disturbances in power quality signals.This model applies attention mechanism together with bi-directional long short-term memory to construct aclassification network and uses Matlab simulation to generate both training and validation data sets.In the new model,seven kinds of common composite disturbance signals were considered,and furthermore they were used as sequence data to di-rectly feed into the network for training and verification.Experiments show that the proposed model could accurately classify disturbance signals and the recognition accuracy of the model on the validation set could reach 99.7%.It is also found that the present model outperformed the conventional machine learning algorithms such as support vector machine and multi-layer perception.