Abstract: To achieve fire early warning and accurately locate the fire source for FPSO units, a real-time monitoring fire warning method based on adaptive neural network was developed, in view of slow convergence and easy to trap in the local extreme points and other issues for the traditional neural network in the fire warning. For developing the method, the traditional neural network was improved by adding momentum and adaptive learning rate, and the network model was trained according to FPSO fire accident data. The fire situation and location were predicted according to the real-time temperature monitoring data. Then, taking process configuration module I area of FPSO platform as an example, an adaptive neural network model was built for real-time monitoring fire warning. A fire scene was set by using the FLACS, and the temperature monitoring data after fire were input into the real-time monitoring fire warning model. The results show that the output fire source coordinates of the fire warning model drop in the combustion area simulated by FLACS.

Key words: floating production storage and offloading units(FPSO), fire source localization, fire warning, adaptive neural network, adaptive learning rate