基于自然语言处理的电力调度语音识别方法  

Speech Recognition Method for Power DispatchingBased on Natural Language Processing

作　　者：胡州明 唐冬来 李玉 朱海萍 宋卫平 颜涛 HU Zhouming;TANG Donglai;LI Yu;ZHU Haiping;SONG Weiping;YAN Tao(Aostar Information Technology Co.,Ltd.,Chengdu 610095,China)

机构地区：[1]四川中电启明星信息技术有限公司,四川成都610095

出　　处：《微型电脑应用》2023年第6期171-174,共4页Microcomputer Applications

摘　　要：为了解决新能源电站调度中存在语音交流处理能力不足、语音识别准确率低的问题,提出了一种基于自然语言处理的电力虚拟调度方法。通过动态时间规整算法提取新能源场站调度语音中的特征信息;采用自然语言处理将调度语音的特征信息进行预处理,降低高维语音空间识别难度,并提取语音文本的词袋特征。在此基础上,通过多标签分类进行上下文关联分析,获得新能源场站调度语音的文本信息。在某省的新能源场站调度中应用所提方法,其在单条调度语言处理时长和语音识别准确率方面较隐马尔科夫方法低0.058 s和4.66%。应用结果表明,所提方法优于隐马尔科夫方法,验证了所提方法的有效性。In order to solve the problems of insufficient speech communication processing capacity and low speech recognition accuracy in the dispatching of new energy power plants,a power virtual dispatching method based on natural language processing is proposed in this paper.The characteristic information in the dispatching voice of new energy stations is extracted by dynamic time warping algorithm.Secondly,natural language processing is used to preprocess the feature information of scheduled speech,reduce the difficulty of high-dimensional speech space recognition,and extract the word bag feature of speech text.On this basis,the voice text information of new energy station dispatching voice is obtained by multi label classification and context correlation analysis.The proposed method is applied to the scheduling of new energy stations in a province,which is 0.058 s and 4.66%lower than the hidden Markov method in terms of single scheduling language processing time and speech recognition accuracy.The application results show that the proposed method is better than hidden Markov method.

关 键 词：电力调度 语音识别 自认语言处理 词袋特征 

分 类 号：TM769[电气工程—电力系统及自动化]