基于双通道融合网络的电力故障报修分类模型

Classification Model of Power Fault Report Based on Dual-channel Fusion Network

针对目前电力故障报修工单分类准确率不高,而静态词向量无法表示多义词,以及传统深度学习模型提取特征不够全面等问题,提出了基于双通道融合网络的电力故障报修分类模型(RoBERTa-BA-MCNN)。RoBERTa预训练模型通过参考上下文语境,得到词的动态语义表示,解决一词多义问题。由BiSRU-Attention模块提取文本上下文序列特征,而多尺度卷积神经网络(MCNN)模型捕获多尺度语句级别的局部特征,将双通道特征拼接融合,得到更为全面的高维特征。在真实电力故障报修数据上进行实验,结果表明,RoBERTa-BA-MCNN模型准确率达到了95.67%,高于实验对比的其他模型。

To address the problems that the classification accuracy of power fault report order is low,the static word vector cannot solve polysemy,and the traditional deep learning model is not comprehensive enough to extract features,a power fault report classification model based on dual-channel fusion network(RoBERTa-BA-MCNN)is proposed.In order to solve polysemy,the dynamic semantic representation of a word is obtained by using pre-trained RoBERTa combined with context.The BiSRU-Attention module extracts the text context sequence features,and the multi-scale convolutional neural network model captures the local features at the multi-scale statement level,and fuses the dual channel features to obtain more comprehensive high-dimensional features.Experiments are carried out on real power fault report data,and the results show that the accuracy of RoBERTa-BA-MCNN reaches 95.67%,which is higher than other models.