基于多传感器信息融合的智能建筑火灾探测研究

发布时间：2018-08-06 20:48

【摘要】：在当代社会中，随着科技和经济的不断发展，商业区、住宅区以及校园区等集中区域越来越多，火灾危害的程度越来越高。如果人们能在火灾初期就能及时发现并且将其迅速扑灭，那么就能够使火灾中的损失降到最低。因此，火灾探测技术在建筑消防系统中的作用显得很重要，对其研究也是有很大意义的。传统的火灾探测系统一般只用到单一探测器，利用其阈值进行火灾信息的判断，然而，现在的建筑越来越复杂，使得探测环境变量以及干扰信号越来越多，单一探测器无法可靠地探测火灾信息。因此，为了提高火灾探测器探测信息的准确性，目前很多科研工作者正在研制多传感火灾探测器来取代传统的火灾探测器。本文结合火灾探测技术的基本特点，建立起基于多传感器信息融合的火灾探测系统的模型。整个系统的探测算法分为三层，即像素层、特征层和决策层，在像素层中，探测器对CO气体、温度、烟雾等特征参数进行数据采集和预处理；在特征层中，将像素层中得到的信息利用人工神经网络和线性变换进行融合，实现目标对象特征明火概率、阴燃火概率、非火灾源概率以及火灾概率的识别，为决策层提供直接性判据和间接性判据；在决策层中，根据特征层提供的信息进行判断分析，当特征层提供的信息可以判定火灾情况，则直接输出，反之，则利用模糊逻辑推理判断技术把像素层的数据再次进行融合得到决策结果。在本文的研究中，利用MATLAB仿真软件对厨房干扰信号、我国标准阴燃火以及标准明火三种火灾信息进行了仿真。在特征层仿真研究中，得出了不同神经网络隐含层的误差曲线，，由此确立了隐含层数目为7的BP神经网络。在决策层仿真研究中，设定的模糊规则得出最终的输出结果。仿真结果表明基于信息融合技术的火灾探测系统具有可行性。
[Abstract]:In contemporary society, with the development of science and technology and economy, there are more and more concentrated areas, such as commercial district, residential district and campus area, and the degree of fire hazard is becoming higher and higher. If people can detect and extinguish the fire in time at the beginning of the fire, the damage in the fire can be minimized. Therefore, the role of fire detection technology in building fire protection system is very important, and the study of fire detection technology is of great significance. The traditional fire detection system usually only uses a single detector to judge the fire information by using its threshold value. However, the buildings are becoming more and more complex, which makes the detection of environmental variables and interference signals more and more. A single detector cannot reliably detect fire information. Therefore, in order to improve the accuracy of fire detection information, many researchers are developing multi-sensor fire detectors to replace the traditional fire detectors. Based on the basic characteristics of fire detection technology, a fire detection system model based on multi-sensor information fusion is established in this paper. The detection algorithm of the whole system is divided into three layers: pixel layer, feature layer and decision layer. In the pixel layer, the detector collects and preprocesses the characteristic parameters such as CO gas, temperature, smoke, etc. The information obtained in the pixel layer is fused with artificial neural network and linear transformation to realize the recognition of the target object's characteristic open fire probability, smoldering fire probability, non-fire source probability and fire probability. Direct criterion and indirect criterion are provided for the decision layer. In the decision layer, the information provided by the feature layer is judged and analyzed. When the information provided by the feature layer can determine the fire situation, it is directly output. Then the decision result is obtained by using fuzzy logic reasoning and judgment technology to fuse the pixel layer data again. In this paper, MATLAB simulation software is used to simulate three kinds of fire information, such as kitchen disturbance signal, standard smoldering fire and standard open fire in China. In the characteristic layer simulation, the error curves of the hidden layers of different neural networks are obtained, and the BP neural networks with the number of hidden layers 7 are established. In the study of decision-making simulation, the fuzzy rules are set to get the final output. The simulation results show that the fire detection system based on information fusion technology is feasible.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TU892

【参考文献】