机器学习在电站设备状态分析中的应用  

Application of machine learning in state analysis of power plant equipment

作　　者：李晓东 陈亚鹏 王保营 胡乔艳 赖菲[2] 吴涛[2] 徐创学[2] 薛晗光[2] 何新[2] 王智微[2] 高海东[2] 高林[2] LI Xiaodong;CHEN Yapeng;WANG Baoying;HU Qiaoyan;LAI Fei;WU Tao;XU Chuangxue;XUE Hanguang;HE Xin;WANG Zhiwei;GAO Haidong;GAO Lin(SPIC Zhuhai Hengqin Cogeneration Co.,Ltd.,Zhuhai 519000,China;Xi'an Thermal Power Research Institute Co.,Ltd.,Xi'an 710054,China)

机构地区：[1]中电投珠海横琴热电有限公司,广东珠海519000 [2]西安热工研究院有限公司,陕西西安710054

出　　处：《热力发电》2020年第1期129-133,共5页Thermal Power Generation

摘　　要：本文对电站设备状态分析中学习向量量化(LVQ)神经网络和深度学习算法循环递归长短期记忆(LSTM)神经网络进行了详细分析,利用LSTM神经网络对磨煤机设备进行状态分析,将LSTM神经网络中最后一个隐含层的激励函数设为Softmax函数,其输出值表示设备状态的健康程度及设备可能发生事故的概率,并将LSTM神经网络和LVQ神经网络进行设备状态分析对比。结果表明,利用LSTM神经网络得到的训练模型可以得到设备状态分类更高的准确率,减少在设备状态评判中的漏报率和误报率。The learning vector quantitative(LVQ)neural network and deep learning algorithm cyclic recurrent longterm and short-term memory(LSTM)neural network in state analysis of power station equipment are analyzed in detail.The LSTM neural network is applied to carry out state analysis for the coal mill equipment,the excitation fimction of the last hidden layer in the LSTM neural network is set as Softmax function,and the output value of the LSTM neural network indicates the health degree of the device state and the probability that the device may have an accident.Moreover,the equipment state analysis result obtained by the LSTM neural network is compared with that of the LVQ neural network.The results show that,the training model obtained by the LSTM neutral network can obtain a higher classification accuracy and reduce the rate of missed and false alarm in equipment condition evaluation.

关 键 词：电站设备 机器学习 深度学习 状态分析 磨煤机 LSTM神经网络 LVQ神经网络 

分 类 号：TM73[电气工程—电力系统及自动化]