Abstract: Traditional fire detection algorithms confront such drawbacks as low accuracy and poor adaptability in complex fire detection environments. To deal with those problems, a new algorithm is proposed for fire detection with multi-sensor data fusion by combining fuzzy sets and with Dempster-Shafer theory. The proposed algorithm first estimates the fire state using the flame, smoke and temperature sensors, and then calculates the fuzzy memberships of these sensors according to the given fuzzy membership function. Besides, a method of sensor credibility calculation is introduced to improve the robustness of the detection system, and each membership and the corresponding credibility measured from the sensors will be transferred to the basic probability assignment function (mass function). Finally, evidence theory is applied to integrate the information of multiple measurements within a period. The results suggest that the proposed algorithm overcomes the drawbacks of instability and uncertainty in single-sensor algorithms, and improves the fire detection accuracy as well as the robustness.

Key words: multi-sensor, fire detection, fuzzy theory, D-S evidence theory, data fusion

摘要： 在复杂多变的火灾检测环境中，针对传统火灾检测方式准确率不高，适应性较差的问题。将模糊集合和D-S证据推理有机结合，提出一种新的用于火灾检测的多传感器数据融合的方法。该方法首先利用火焰、烟雾和温度传感器感知火灾状态，然后根据给出模糊隶属函数计算各个传感器的模糊隶属度。为了增强系统的抗干扰性，引入了计算传感器可信度的方法，并根据每次测量隶属度和可信度转化为基本概率分配函数(mass函数)，最后利用证据理论对一个周期内多次测量的信息进行融合。结果表明，该方法提高了火灾检测判别的准确率，克服单个传感器带来的不稳定性和不确定性，增强了火灾检测系统的鲁棒性。

关键词: 多传感器, 火灾检测, 模糊理论, D-S证据理论, 数据融合