Abstract: The complex combustible materials in ancient temple buildings, diverse fire characteristic parameters, and the presence of interference sources such as incense, candles, and smoke have led to high false alarm rates and poor reliability of single parameter fire detection methods. This article proposes a multi parameter composite detection method, which constructs a fire alarm model that adapts to the structure and environment of ancient temple buildings through data collection and fusion of early fire characteristic parameters, thereby achieving accurate detection of early fires in ancient buildings. Represented by the Yonghe Gate Hall, Yongyou Hall, and Wanfu Pavilion of Yonghe Palace, fire characteristic parameters such as gas volume fraction, temperature, and visibility were obtained through numerical simulation. A 4-layer BP neural network was used to train a fire alarm model using a combination of training and testing sets from two dimensions. The results showed that the accuracy of the training alarm model fused with 5 and 3 detection data was similar, with CO volume fraction, temperature, and visibility as sensitive feature parameters; The fire alarm model has strong generalization ability for ancient buildings with similar building structures, with a fire alarm accuracy rate of up to 0.950 0. The fire alarm accuracy rate for ancient buildings with non similar structures is significantly reduced to only 0.720 0; The training model based on independent building fire simulation parameters has a high accuracy rate of 0.930 0 or above for the fire alarm of the building itself. The research process and results provide a theoretical basis for selecting composite fire characteristic parameters, and provide guidance methods and implementation paths for constructing an adaptive fire alarm model with multi detection data fusion for important temple ancient buildings.

Key words: ancient temple buildings, fire detectors, BP neural network, fire characteristic parameters, composite detection technology