Research On Extreme Learning Maching Algorithm For Indoor Oriented Fire Protection

Among various common disasters of modern city, fire can bring heavy damage to the people, goods and materials of indoor spaces of industrial and mining enterprises, residence, tunnel traffic, etc. In recent years, fire detection technology has got certain improvement in veracity, but still can’t meet the demand of high-veracity of construction indoor fire detection.This paper firstly introduced the basic principle and development of extreme learning machine(ELM), and the difference between BP neural network and ELM.Second, this paper introduced a new model selection method of ELM based on multi-objective optimization. This method views ELM model selection as a multi-objective global optimization problem, in which the generalization error and output weights are as optimization objectives. To accelerate the optimization speed, a fast Leave-one-out error estimate of ELM is introduced to refer to the generalization error. Taking into account the contradiction between these two objectives, multi-objective comprehensive learning particle swarm optimization algorithm is utilized to find non-dominated solutions.Finally, aiming at the serious category imbalance of those collected data, and using some existing algorithms cannot recognize fire video data problem rapidly, this paper has come up with the online sequential extreme learning machine algorithm that based on the imbalance of principal curve. This algorithm balances the samples of all kinds and reduces the blindness of less-class samples’ synthetic process on the basis of the distribution character of the online sequence data. This algorithm has solved some difficulty that existing in currently the fire detection field to some extent, and has provided a more effective idea and method for this field.The research in this paper can not only improve the theory and algorithms of Extreme Learning Machine and provide a new solution for online sequential imbalance data classification, but also reduce the damage of fire hazard and implement the real-time monitoring and rapid alert of indoor fire, which demonstrates this research has crucial practical theoretical significance and vital application value.