摘要
基于多层前馈神经网络提出了火灾实验中不同材料热释放率的学习算法和预测技术．同时，将具有全局收敛特性的混合共轭梯度（ＭＣＧ）算法应用于该问题中多层前馈神经网络的训练，克服了传统ＢＰ算法收敛速度慢，推广性能差的缺陷．文中对ＭＣＧ方法进行了大量模拟，并将模拟结果与ＢＰ算法及带有动量项的ＢＰ算法作了全面比较。
A feedforward multilayer neural network based method for calculation and prediction for the rate of heat release (RHR) of different materials in fire safety science is proposed. A global convergent mixed conjugate gradient (MCG) algorithm is also proposed for training our multilayer neural network, which overcomes the slow training speed and poor generalization capability of the traditional BP algorithm. A number of computer simulations show that the proposed NN based method for calculating RHR is very efficient and that the MCG is also superior to the BP algorithm in convergent and generalized properties.
基金
安徽省自然科学基金
关键词
神经网络
混合共轭梯度
热释放率
火灾
火焰
Multilayer Feedforward Neural Network, Learning Algorithm, Mixed Conjugate Gradient, Rate of Heat Release, Global Convergence
