摘要
针对火灾信号的时变性和非线性特点以及单特征信号火灾预测方法的高漏报率和误报率的问题,提出一种基于LSTM(Long Short Time Memory,长短时记忆网络)和RBF-BP(Radial Basis Function and Back Propagation,径向基反向传播网络)深度学习模型的多源信息融合火灾预测方法。首先,在信息层通过多种传感器采集火灾发生时的特征信息,并通过速率检测算法对特征信息进行预处理。然后在特征层利用深度学习中的LSTM和RBF-BP神经网络对火灾多种特征信号进行自适应学习,输出有火、无火以及阴燃火三种火型的发生概率。最后采用模糊逻辑控制系统决策输出有无火灾发生。仿真实验结果表明,所提出的火灾预测方法与传统的和单神经网络方法相比具有更高的预测精度和自适应性。
Aiming at the time-varying and non-linear characteristics of fire signal and the high rate of false alarm and missed alarm of single feature signal fire prediction method,a multi-source information fusion fire prediction method based on LSTM(Long Short Time Memory)and RBF-BP(Radial Basis Function and Back Propagation)deep learning model is proposed.Firstly,the characteristic information of fire is collected by many kinds of sensors in the information layer,and the characteristic information is preprocessed by rate detection algorithm.Then,in the feature layer,the LSTM and RBF-BP neural network in the deep learning are used to learn the fire feature signals adaptively,and the occurrence probability of fire,no fire and smoldering fire are output.Finally,the fuzzy logic control system is used to decide whether there is fire or not.The simulation results show that the proposed method is more accurate and adaptive than the traditional and single neural network methods.
作者
许春芳
乔元健
李军
XU Chun-fang;QIAO Yuan-jian;LI Jun(Lingcheng District Bureau of water resources of Dezhou City,Dezhou 253500,China;School of Electrical Engineering and Automation,Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China;School of Electronic Information Engineering(University Physics Teaching Department),Qilu University of Technology(Shandong Academy of Sciences),Jinan 250353,China)
出处
《齐鲁工业大学学报》
2020年第3期53-59,共7页
Journal of Qilu University of Technology
基金
山东省研究生教育计划创新项目(SDYY16032)。
作者简介
许春芳,副高级工程师,研究方向:水利工程建设管理;通讯作者:李军,博士、副教授,研究方向:深度学习、MIMO-OFDM、协作通信、认知无线电,Email:rogerjunli@sdu.edu.cn。
