基于极限学习机的炭化可燃物着火时间预测

机器学习技术近年来在许多传统科学领域取得了应用,针对火灾中炭化可燃物着火时间与物性参数及环境参数之间关系复杂的特点,提出了一种基于极限学习机的预测方法以实现不同物性及环境参数时着火时间的快速准确预测,为防治及扑救火灾提供参考。首先建立炭化可燃物热解数值模型,考虑了可燃物热解过程中的含水率以及热解反应、气体流动等复杂物理化学反应过程,然后搭建极限学习机,以数值模拟数据为基础进行训练及验证工作。结果表明基于极限学习机的预测方法能够有效实现炭化可燃物着火时间的快速准确预测,平均相对误差小于3%。

Predicting ignition delay time of charring combustible by extreme learning machine

Machine learning has attracted much attention in various areas in recent years. In this paper, we report an approach to make accurate and fast prediction of the ignition delay time of charring combustibles with different parameters by extreme machine learning. A numerical model taking account of the moisture and various physical and chemical processes including pyrolysis and gas flow is first developed to generate a database for the dependence of ignition delay time on parameters. Extreme learning machine is then employed to train the neural network with the database. Random parameters are finally input into the network to validate the method. Results suggest that the learning-based method could effectively predict the ignition delay time of charring combustibles with an average error less than 3%.