典型舱室火灾智能识别的贝叶斯网络模型

发布时间：2018-11-11 12:27

【摘要】：基于先进传感器,建立了火灾大小和类型智能识别的贝叶斯网络模型,上层温度、下层温度、CO浓度、CO2浓度、O2浓度和遮光度等6个火灾特征参数为识别模型的输入变量,火灾大小和类型为输出变量,并推导了输入变量与输出变量之间的关系。分别在住舱、指挥室、机舱和机库等4种典型舱室模拟了床垫火、电缆火、油池火和喷射火等4种火源,利用CFAST软件得到了2 880组模拟样本数据,对模型参数进行了训练,并根据全尺度火灾试验数据对训练后的识别模型进行了验证。验证结果表明:在火灾传感器数据完整时,对小火、中火和大火状态的平均识别正确率分别为88.0%、95.0%、85.7%,对固体火和油料火的平均识别正确率分别为90.2%、81.5%;在火灾损害严重或武器打击致使单个传感器失效的情况下,对火灾大小和类型的平均识别正确率分别为82.4%、82.7%,比火灾传感器数据完整时分别降低8.1%、2.8%。可见,识别模型具有良好的识别能力和鲁棒性,可应用于舰船损管监控系统,为指挥员选择最有效的灭火方法和战术提供实时的决策支持。
[Abstract]:Based on advanced sensors, a Bayesian network model for intelligent identification of fire size and type is established. Six fire characteristic parameters, such as upper temperature, lower temperature, CO concentration, CO2 concentration, O2 concentration and shading degree, are the input variables of the model. The size and type of fire are output variables, and the relationship between input variables and output variables is derived. Four kinds of typical compartments, such as mattress fire, cable fire, oil pool fire and jet fire, were simulated in four typical compartments of cabin, command room, engine room and hangar, respectively. 2880 groups of simulated sample data were obtained by CFAST software, and the model parameters were trained. The training model is verified based on the full scale fire test data. The results show that when the data of the fire sensor is complete, the average recognition accuracy of small fire, moderate fire and fire state is 88.0 and 95.057, respectively. The average recognition accuracy of solid fire and oil fire is 90.2% and 81.5% respectively. In the case of serious fire damage or the failure of a single sensor caused by a weapon strike, the average correct recognition rate for the size and type of the fire is 82.4 and 82.7, respectively, which is 8.1 percent lower than when the data of the fire sensor is complete. 2.8. It can be seen that the recognition model has good recognition ability and robustness. It can be applied to ship damage management monitoring system and provide real-time decision support for commanders to choose the most effective fire extinguishing methods and tactics.
【作者单位】：大连理工大学船舶工程学院;海军大连舰艇学院航海系;海军装备研究院标准规范研究所;
【基金】：“十二五”国防预研项目(4010403010208) 武器装备军内科研重点项目(2014HJ0030)
【分类号】：U664.88

【相似文献】