基于深度学习的新型电力智能交互平台多任务集成模型研究  

Research on multi-task ensemble model based on deep learning for novel power intelligent interaction platform

作　　者：程超 葛维 郭兰柯 陈博 张亚炜 Cheng Chao;Ge Wei;Guo Lanke;Chen Bo;Zhang Yawei(State Grid Hebei Electric Power Company,Shijiazhuang 050000,China;Shijiazhuang Power Supply Company,State Grid Hebei Electric Power Company,Shijiazhuang 050000,China;Marketing Service Center,State Grid Hebei Electric Power Company,Shijiazhuang 050000,China)

机构地区：[1]国网河北省电力有限公司,石家庄050000 [2]国网河北省电力有限公司石家庄供电分公司,石家庄050000 [3]国网河北省电力有限公司营销服务中心,石家庄050000

出　　处：《电测与仪表》2023年第6期81-85,109,共6页Electrical Measurement & Instrumentation

基　　金：国网河北省电力公司科技项目(TSS2018-03)。

摘　　要：意图理解是新一代电力智能交互平台中的一项基础技术。通过将用户诉求自动分类与分级,可以大幅提升服务效率和质量。针对电力交互平台中的意图理解问题,提出一种基于深度学习的多任务集成模型,该模型可以同时训练意图理解中密切相关的两项子任务:意图检测(Intent Detection)与语义槽填充(Slot Filling)。使用具有长短期记忆(Long-Short Term Memory,LSTM)结构和门控循环单元(Gated Recurrent Unit,GRU)的深度双向循环神经网络(Recurrent Neural Network,RNN)作为基本分类器,多层感知机(Multi-Layer Perceptron,MLP)框架用于组合输出结果,并基于词向量特征与词性特征对模型进行增强。在真实数据上的实验表明该集成多任务模型相比单一模型或其他主流方法更为有效。Intent understanding is a fundamental technology in novel power intelligent interaction platform.Through classifying and grading intents of customers automatically,the efficiency and quality of power service can be remarkably improved.Towards intent understanding problem in power interaction platform,a multi-task ensemble model based on deep learning is proposed,which can simultaneously train two closely related sub-tasks in intent understanding,named intent detection and slot filling.Recurrent neural network(RNN)with long-short term memory(LSTM)and gated recurrent unit(GRU)respectively are used as basic classifier in the proposed model,and multi-layer perceptron(MLP)generates the final output.Word vectors and part-of-speech(POS)features are used to reinforce the proposed model.Experimental results on real-world dialogue data indicates the superiority of the proposed multi-task ensemble model compared with independent models and other peer models.

关 键 词：意图理解 深度学习 循环神经网络 自然语言处理 电力服务 

分 类 号：TM93[电气工程—电力电子与电力传动]