【摘要】 介绍了一种多层局部回归神经网络,并用该网络建立烧结过程的预测模型,提出一种新的学习算法:自适应变步长动态梯度下降法,从而加速了收敛过程.文中,同时研究了网络的抗噪能力,并对其动态建模能力与自适应延时神经网络(ATNN)进行了比较.仿真结果表明:该网络具有较强的抗噪能力,且具有很强的动态建模能力,因而适用于复杂生产过程的建模、预测与控制.更多还原

【Abstract】 A kind of multilayer local recurrent neural networks is introduced to build the predicting model for sinter process. The adaptive step dynamical gradient descent learning algorithm is used to train the network. So, the convergence speed is improved. The experiment shows that: This neural network is robust, to noise and is powerful in modeling and predicting. Therefore, it’s suitable for modeling, predicting and control of complex production process.更多还原