BP Neural Network-based High-rise Building Fire Assessment

With China’s rapid economic development, especially in large and medium-sizedcities, due to rapid population growth, shortage of land resources, high-rise building bya large number of applications in modern society, and gradually become the mainstream.However, the proportion of high-rise building fire in a building fire is high, theprobability of the fire. Loss of large, surprising. Therefore, in recent years, high-risebuilding fire safety has been more and more attention.In this paper, high-rise building fire hazard analysis of factors affecting thehigh-rise building fire safety into the fire protection systems, fire extinguishing systems,security, evacuation systems, fire management four factors. Based on the systematicprinciple continuity principle, the particularity of the principle of universality of theevaluation index, the principle of evaluation indicators can be quantified fire factor isdivided into a number of small factors, and analysis. Established based on BP neuralnetwork high-rise building fire safety evaluation system, and in accordance withnational standards will affect the high-rise building fire safety factors is divided into fivegrades. BP neural network input layer, hidden layer, the design and tuning of the outputlayer, the establishment of the BP neural network model of high-rise building fire safety.According to the high-rise building-specific data and expert judgment, collected includehigh-rise building fire protection systems, fire extinguishing systems, securityassessment sample of the safe evacuation systems, management of several factors.The training samples from Baoding Fire Station. The use Matlab7.0realization ofthe simulation on high-rise building anti-fire assessment, the use of plus momentumfactor and variable step size optimization algorithm to solve the defects of slowconvergence speed of BP neural network modeling and simulation process. Continuouscycle of learning and training, and constantly adjust the weights and thresholds of thelayers in the network model, and finally get the ideal model, and into the test sample totest calibration of the model. Practice has proved that the final output error to achievethe accuracy requirements with the model, the model has little practical value, but stillneeds further improvement and complete.