Abstract: Aiming at the shortcomings of missing alarms, false alarms in single sensor prediction, this paper proposes a MK-XGBoost fire detection technology based on multi-sensor data fusion. By collecting the temperature, smoke concentration and CO concentration in a confined space, a trend factor is generated based on the Mann-Kendall method. The factor is positively related to the intensity of the upward trend. Then the fire data and the trend factor are input into the XGBoost algorithm as features to determine if a fire occurred. The simulation was carried out under the software FDS and MATLAB. Compared with the original algorithms of SVM and XGBoost, the simulation results show that the accuracy of fire detection is 98.0%, and the recognition time is increased by 0.9 s. Therefore, the MK-XGBoost algorithm can effectively improve the accuracy of fire recognition.

Key words: multi -sensor fusion; fire recognition; XGBoost; Mann-Kendall