Teng, WanxiuWanxiuTengLiu, YantongYantongLiuLi, QiangQiangLiXu, JinlongJinlongXuTian, XinXinTian2025-11-272025-11-272025-04-02International Conference on Artificial Intelligence and Autonomous Transportation, AIAT 2024https://hdl.handle.net/11420/59232The smoke detection technology has high false alarm rate and is greatly affected by ambient wind speed. It is difficult for the temperature detector to realize early alarm, and there are risks of missing alarm and delayed alarm. Composite fire detection technology is an effective means to solve the above problems. In the application of composite fire detection algorithm, there are some problems such as false positives in frame-by-frame state judgment, delay alarm and fuzzy alarm time caused by unclear setting of detection interval length. In this paper, 7 standard combustibles in the equipment compartment under vehicle were tested on the fire detection simulation test platform, and typical fire characteristic parameters such as ambient temperature, CO concentration, VOC concentration and smoke concentration were obtained under 3~6 m/s horizontal wind field. Through ensemble learning and LSTM algorithm, the following conclusions are obtained: The detection interval length with the highest accuracy of state recognition is 30~40 s; The state determination algorithm model based on Xgboost - LSTM algorithm is constructed. Increased to 92.56%, the model can efficiently learn sample experimental data and accurately judge the environmental state; Within the detection interval length, the accurate alarm rate of smoulder is 78%, the accurate alarm rate of open flame is 96%, the false alarm rate and false alarm rate are 4% and 2%. The wavelet transform algorithm can judge the fire state within the detection interval length and achieve accurate alarm.enAlgorithmEquipment Compartment under VehicleFire DetectionTechnology::620: EngineeringConference Paper10.1007/978-981-96-3977-9_20Conference Paper