An advanced fire estimation model for decentralized building control

This paper describes the development of an inverse model to identify the fire location and intensity level based on sensor data and physical-model output for use with a decentralized control system. The model was proven to be accurate and effective. The inverse procedure can be divided into two steps. In the first step, the fire estimation is conducted in pair-wise adjacent zones based on Bayesian inference. A local relative probability ratio is calculated. In the second step, the global inference is derived from the cooperation of all zones. For demonstration purposes, the model was applied to a decentralized control system with limited information of the building structure and sensor readings. Based on the building layout, the fire location and intensity level can be determined effectively. The results from the decentralized control are compared with those obtained for conventional centralized control with full information of the building structure and sensor readings. The decentralized fire estimation model proved to be accurate for practical applications. The results were also found to be insensitive to the traversal path selection in the topology of zones.