-
随着社会发展,城市人口愈发密集,火灾发生数量逐年增加。火灾造成财产损失和人员伤亡,给社会带来不可估量的伤害,因此如何及时准确地探测火灾至关重要。
传统检测火灾的传感器,使用了烟雾探测器。烟雾探测器根据探测原理可分为两类:基于电离的探测器和基于光电的探测器。基于光电的烟雾探测器利用烟雾粒子影响光的散射来探测烟雾;基于电离的烟雾探测器则利用烟雾颗粒的存在会导致电极之间的电流下降来检测烟雾。但是单一信号判别火灾易受到环境影响,产生虚警,准确率不高。为了提升火灾探测准确率,需要使用多传感器探测火灾,因此选取合适的探测数据源、合理融合多传感器信息成为了探测火灾的关键。
目前用于判别火灾的数据源多来自CO浓度、CO2浓度、烟雾浓度、温湿度和光照强度。而融合多传感器信息方法则多种多样。如文献[1-2]使用模糊逻辑融合温度、火焰强度、烟雾浓度信息;文献[3]使用模糊逻辑融合烟雾、温度、湿度、CO浓度信息;文献[4]先使用卡尔曼滤波对温度、湿度和烟雾信息进行预处理,再通过模糊逻辑判别室外火灾。模糊逻辑虽然可以模拟人的思考判断方式,但是模糊规则多由人为经验确定,不具有客观性。文献[5-6]使用反向传播(back propagation, BP)神经网络训练CO、温度和烟雾来提升火灾探测准确率;文献[7]使用卷积神经网络(convolutional neural networks, CNN)训练数据检测火灾,但是神经网络需要大量的训练数据,而火灾数据不易采集。文献[8-14]提出了基于图像、视频处理等方法检测火灾的方法,但是计算资源消耗过大,所用时间较长。文献[15]使用D-S证据理论融合温度、烟雾和光照强度信息,并且规定冲突因子K小于阈值来判别火灾;文献[16]则单纯使用D-S证据理论融合温度和烟雾浓度,但是D-S证据理论没有充分考虑证据之间的相互关系,容易造成信任悖论,产生不可能发生的情况或者与事实相反的情况。目前对于D-S证据理论的改进,可分为3类:1) 为数据源分配权重,对其改进[17-19];2) 对合成规则的改进[20-21];3) 既修改证据源,也改进合成规则[22]。
本文使用温度、烟雾浓度、CO浓度、O2浓度、热释放速率作为数据源,使用火灾判别概率函数计算每个数据源对应的火灾判别概率向量P,结合Jousselme距离为各个数据源分配权重,最后通过D-S证据理论对多传感器判别信息融合,得到最终的火灾判别概率。本方法充分考虑了各个信号之间的关系,仿真结果表明,在保证准确率的基础上,可提前检测出火灾,有效提升火灾探测及时性。
-
本文将火灾结果分为无火、阴燃火、明火3种,分别记为A,B,C。即识别框架为
$\theta = \left\{ {A,B,C} \right\}$ 。假设一个函数M满足如下条件:$$ M:{2^U} \to \left[ {0,1} \right] $$ (1) $$ M\left( \phi \right) = 0$$ (2) $$ \sum\limits_A {M\left( I \right)} = 1 $$ (3) 则称M为
$\theta $ 上的基本概率分配函数,$M\left( I \right)$ 称为I的基本概率值,其中$\phi$ 代表空集。将各个数据源通过火灾判别概率函数所得的各自火灾判别概率向量P作为其对应基本概率分配函数。对于
$\forall I \subset \theta$ ,识别框架$\theta $ 上的有限个基本概率分配函数$m_1,m_2,\cdots,m_n$ 的合成规则为:$$ \begin{split}& \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\left( {{m_1} \oplus {m_2} \oplus \cdots \oplus {m_5}} \right)\left( I \right) = \hfill \\ &\frac{1}{K}\sum\nolimits_{{I_1} \cap {I_2} \cap \cdots {I_5} = I} {{m_1}\left( {{I_1}} \right){m_2}\left( {{I_2}} \right) \cdots {m_n}\left( {{I_5}} \right)} \hfill \\ \end{split} $$ (4) 式中,
$K$ 为归一化因子,计算公式如下:$$ K = \sum\nolimits_{{I_1} \cap {I_2} \cap \cdots {I_5} \ne \phi } {{m_1}\left( {{I_1}} \right){m_2}\left( {{I_2}} \right) \cdots {m_n}\left( {{I_5}} \right)} $$ (5) -
针对经典D-S证据理论没有考虑证据之间的差异,引起信任悖论的现象,本文使用Jousselme距离[24]评估证据源之间差异,进而为各个证据源分配权重。
首先,构建距离矩阵DM如下:
$$ {\bf{DM}} = \left( {\begin{array}{*{20}{c}} 0&{{d_{12}}}&{{d_{13}}}&{{d_{14}}}&{{d_{15}}} \\ {{d_{21}}}&0&{{d_{23}}}&{{d_{24}}}&{{d_{25}}} \\ {{d_{31}}}&{{d_{32}}}&0&{{d_{34}}}&{{d_{35}}} \\ {{d_{41}}}&{{d_{42}}}&{{d_{43}}}&0&{{d_{45}}} \\ {{d_{51}}}&{{d_{52}}}&{{d_{53}}}&{{d_{54}}}&0 \end{array}} \right) $$ (6) 式中,矩阵元素
$ {d_{ij}} $ 定义如下:$$ {d_{ij}} = d\left( {{{{\boldsymbol{m}}_i}} , {{{\boldsymbol{m}}_j}} } \right) = \sqrt {\frac{1}{2}{{\left( { {{{\boldsymbol{m}}_i}} - {{{\boldsymbol{m}}_j}} } \right)}^{\rm{T}}}{\boldsymbol{D}}\left( {{{{\boldsymbol{m}}_i}} - {{{\boldsymbol{m}}_j}} } \right)} $$ (7) ${{{\boldsymbol{m}}_i}}$ 和${{{\boldsymbol{m}}_j}}$ 代表某个数据源当前时刻的火灾判别概率向量;$ {d_{ij}} $ 代表${{{\boldsymbol{m}}_i}}$ 和${{{\boldsymbol{m}}_j}}$ 之间的Jousselme距离;D是一个${2^N} \times {2^N}$ 的矩阵,矩阵元素为${\boldsymbol{D}}(A,B) = $ $ \dfrac{{\left| {A \cap \left. B \right|} \right.}}{{\left| {A \cup \left. B \right|} \right.}}$ ,式(7)可简化为:$$ \begin{split}& \;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;\;{d_{ij}} = d\left( { {{{\boldsymbol{m}}_i}} , {{{\boldsymbol{m}}_j}} } \right) \hfill = \\ & \sqrt {\frac{1}{2}\left[ {\left( {{{\left\| {{{{\boldsymbol{m}}_i}} } \right\|}^2}} \right) + \left( {{{\left\| { {{{\boldsymbol{m}}_j}} } \right\|}^2}} \right) - 2\left\langle { {{{\boldsymbol{m}}_i}} ,{{{\boldsymbol{m}}_j}} } \right\rangle } \right]} \hfill \\ \end{split} $$ (8) 式中
$\left\langle { {{{{\boldsymbol{m}}_i}} , {{{\boldsymbol{m}}_j}} }} \right\rangle$ 为向量${{{\boldsymbol{m}}}_{i}}、{{{\boldsymbol{m}}}_{j}}$ 的内积。其次,由式(6)计算某个数据源对应的P所占的权重,公式如下:
$$ {u_{ij}} = {e^{ - {d_{ij}}}} $$ (9) $$ {\alpha _i} = \frac{{\displaystyle\sum\limits_{j = 1}^5 {{u_{ij}}} }}{{\displaystyle\sum\limits_{i = 1}^5 {\displaystyle\sum\limits_{j = 1}^5 {{u_{ij}}} } }} $$ (10) 式中,
${\alpha _i}$ 是矩阵DM第i行数据源判别概率向量所占的权重。再根据每个证据源的权重,修改证据源,公式如下:
$$ {{\boldsymbol{m}}'} = \sum\limits_{i = 1}^5 {{\alpha _i}{{{\boldsymbol{m}}_i}} } $$ (11) 式中,
${{{\boldsymbol{m}}_i}}$ 是每个数据源对应的判别概率向量;${{\boldsymbol{m}}'}$ 是各个证据源融合后的判别概率向量。最后,对
${{\boldsymbol{m}}'}$ 进行n−1次D-S证据理论融合,得到最终的判别概率 ${\boldsymbol{P}}\left( {{I_i}} \right)$ 。式(4)可简化为:$$ \begin{split}& {\boldsymbol{P}}\left( {{I_i}} \right) = \left( { {{\boldsymbol{m}}'} \oplus {{\boldsymbol{m}}'} \oplus {{\boldsymbol{m}}'} \oplus {{\boldsymbol{m}}'} \oplus {{\boldsymbol{m}}'} } \right)\left( {{I_i}} \right) = \frac{1}{K}\times \hfill \\ &\;\;\left[ { {{\boldsymbol{m}}'} \left( {{I_i}} \right)* {{\boldsymbol{m}}'} \left( {{I_i}} \right)* {{\boldsymbol{m}}'} \left( {{I_i}} \right)* {{\boldsymbol{m}}'} \left( {{I_i}} \right)* {{\boldsymbol{m}}'} \left( {{I_i}} \right)} \right] \hfill \\ \end{split} $$ (12) 式(5)可简化为:
$$ K = \sum\limits_{i = 1}^3 {[m'(I_i} )]^5 $$ (13) 式中,
${I_i} \subset \theta$ 。由式(12)和式(13)计算阴燃火概率和明火概率,将其相加作为火灾发生概率${\boldsymbol{P}}\left( {{\rm{fire}}} \right)$ ,即:$$ {\boldsymbol{P}}\left( {{\rm{fire}}} \right) = {\boldsymbol{P}}\left( B \right) + {\boldsymbol{P}}\left( C \right) $$ (14) 具体算法如图4所示,过程如下:1)获取5种火灾信号:CO浓度、烟雾浓度、温度、O2浓度差值和热释放速率,作为数据源;2) 将数据源通过 图3火灾判决概率函数,分别得到各个数据源的火灾判别概率;3) 构建距离矩阵,通过式(6)~式(11)计算各个数据源权重并且修改证据源;4) 融合后的证据源进行4次证据理论融合,得到最终火灾发生概率;5) 将阴燃火和明火概率相加作为火灾发生概率,若大于0.8,则判定为火灾发生。
D-S Fusion Detection Method with New Data Sources
-
摘要: 针对火灾检测时延过长的问题,该文引入新的火灾探测数据源,将模糊逻辑和D-S证据理论融合,提出一种信号火灾探测方法。该方法使用CO浓度、烟雾浓度、温度、O2浓度以及热释放速率等作为火灾探测数据源,建立火灾判别概率函数,计算各个数据源的无火、阴燃火和明火的判别概率,结合Jousselme距离为数据源分配权重,最终通过D-S证据理论对多源判别信息进行融合。仿真结果表明,该方法相比于未引入O2浓度和热释放速率的火灾探测方法,能提早3~5 s探测出火灾,提升了火灾探测及时性。Abstract: Aiming at the problem of excessive fire detection time delay, a new data source of fire detection is introduced. Fuzzy logic and D-S evidence theory are merged, and a signal fire detection method is proposed. This method uses CO concentration, smoke concentration, temperature, O2 concentration, and heat release rate as fire detection data sources. It establishes a fire discrimination probability function to calculate the discrimination probability of no fire, smoldering fire, and open flame for each data source. The data sources are assigned weights using the Jousselme distance, and the multi-source discriminated information is fused through D-S evidence theory. The simulation results show that this method can detect fire 3-5 seconds earlier compared with the fire detection method that does not introduce O2 concentration and heat release rate, thus improving the timeliness of fire detection.
-
Key words:
- data fusion /
- D-S evidence theory /
- fire detection /
- fuzzy theory /
- multi-sensor
-
[1] SOWAH R, OFOLI A, KRAKANI S, et al. Hardware module design of a real-time multi-sensor fire detection and notification system using fuzzy logic[C]//IEEE Industry Applications Society Meeting. Vancouver, BC: IEEE, 2014: 1-6. [2] SOWAH R, OFOLI A, KRAKANI S, et al. Hardware design and web-based communication modules of a real-time multisensorfire detection and notification system using fuzzy logic[J]. Industry Applications, IEEE Transactions on, 2017, 53(1): 559-566. doi: 10.1109/TIA.2016.2613075 [3] GIANDIO, SARNO R. Prototype of fire symptom detection system[C]//International Conference on Information and Communications Technology (ICOIACT). Yogyakarta: IEEE, 2018: 489-494. [4] YODDUMNERN A, YOOYATIVONG T, CHAISRICHAROEN R. The wifi multi-sensor network for fire detection mechanism using fuzzy logic with IFTTT process based on cloud[C]//2017 14th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON). Phuket: IEEE, 2017: 785-789. [5] LIANG Y H, TIAN W M. Multi-sensor fusion approach for fire alarm using BP neural network[C]//International Conference on Intelligent Networking & Collaborative Systems. [S.l.]: IEEE, 2016: 99-102. [6] CHENG C X, SUN F C, ZHOU X Q. One fire detection method using neural networks[J]. Tsinghua Science & Technology, 2011, 16(1): 31-35. [7] LUO Y, ZHAO L, LIU P, et al. Fire smoke detection algorithm based on motion characteristic and convolutional neural networks[J]. Multimedia Tools and Applications, 2018, 77: 15075-15092. doi: 10.1007/s11042-017-5090-2 [8] GRAHAM C. Detecting the right technology[J]. International Fire Protection, 2010, 42(5): 39-45. [9] LI G D, LU G, YAN Y. Fire detection using stereoscopic imaging and image processing techniques[C]//IEEE International Conference on Imaging Systems and Techniques (IST2014). Santorini: IEEE, 2014: 28-32. [10] MEMARZADEH B, MOHAMMADI M A. Fire detection using multi criteria image processing technique in video sequences[J]. Indonesian Journal of Electrical Engineering and Computer Science, 2015, 16: 136-144. [11] KHATAMI A, MIRGHASEMI S, KHOSRAVI A, et al. A new PSO-based approach to fire flame detection using K-Medoids clustering[J]. Expert Systems with Applications, 2017, 68(C): 69-80. [12] MUHAMMAD K, AHMAD J, LV Z, et al. Efficient deep CNN-based fire detection and localization in video surveillance applications[J]. IEEE Transactions on Systems, Man, and Cybernetics:Systems, 2019, 49(7): 1419-1434. doi: 10.1109/TSMC.2018.2830099 [13] CHEN K, CHENG Y, BAI H, et al. Research on image fire detection based on support vector machine[C]//2019 9th International Conference on Fire Science and Fire Protection Engineering (ICFSFPE). Chengdu: [s.n.], 2019: 1-7. [14] HUANG H Y, KUANG P, FAN L, et al. An improved multi-scale fire detection method based on convolutional neural network[C]//2020 17th International Computer Conference on Wavelet Active Media Technology and Information Processing. Chengdu: [s.n.], 2020: 109-112. [15] SADEWA R P, IRAWAN B, SETIANINGSIH C. Fire detection using image processing techniques with convolutional neural networks[C]//2019 International Seminar on Research of Information Technology and Intelligent Systems (ISRITI). Yogyakarta: IEEE, 2019: 290-295. [16] TING Y Y, HSIAO C W, WANG H S. A data fusion-based fire detection system[J]. IEICE Transactions on Information and Systems, 2018, 101(4): 977-984. [17] ZHANG Q, CAI Z, CHEN M. Application of data fusion technology based on D-S evidence theory in fire detection[C]//Sixth International Conference on Electronics and Information Engineering. Dalian: SPIE, 2015: [18] MURPHY C K. Combining belief functions when evidence conflicts[J]. Decision Support System, 2000, 29(1): 1-9. doi: 10.1016/S0167-9236(99)00084-6 [19] 邓勇, 施文康, 朱振福. 一种有效处理证据冲突的组合方法[J]. 红外与毫米波学报, 2004, 23(1): 27-32. doi: 10.3321/j.issn:1001-9014.2004.01.006 DENG Y, SHI W K, ZHU Z F. Efficient combination approach of conflict evidence[J]. Journal of Infrared and Millimeter Waves, 2004, 23(1): 27-32. doi: 10.3321/j.issn:1001-9014.2004.01.006 [20] LI K, HAN Y. An evidence fusion method based on weight optimization[C]//2017 36th Chinese Control Conference (CCC). Dalian: IEEE, 2017: 5089-5093. [21] YAGER R R. On the dempster-shafer framework and new combination rules[J]. Information Sciences, 1987, 41(2): 93-137. doi: 10.1016/0020-0255(87)90007-7 [22] SMETS P. The combination of evidence in the transferable belief model[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1990, 12(5): 447-458. doi: 10.1109/34.55104 [23] 卢正才, 覃征. 证据合成的一般框架及高度冲突证据合成方法[J]. 清华大学学报, 2011(11): 1701-1705. LU Z C, QIN Z. General framework for evidence combination and its approach to highly conflicting evidence fusion[J]. Journal of Tsinghua University, 2011(11): 1701-1705. [24] 肖建于, 童敏明, 朱昌杰, 等. 基于广义三角模糊数的基本概率赋值构造方法[J]. 仪器仪表学报, 2012, 33(2): 429-434. doi: 10.3969/j.issn.0254-3087.2012.02.027 XIAO J Y, TONG M M, ZHU C G, et al. Basic probability assignment construction method based on generalized triangular fuzzy number[J]. Chinese Journal of Scientific Instrument, 2012, 33(2): 429-434. doi: 10.3969/j.issn.0254-3087.2012.02.027