基于多层局部回归神经网络的复杂生产过程预测模型  

A MULTILAYER LOCAL RECURRENT NEURAL NETWORK BASED PREDICTING MODEL FOR COMPLEX PRODUCTION PROCESS

作　　者：张兰玲[1] 刘贺平[1] 孙一康[1] 

机构地区：[1]北京科技大学自动化系,北京100083

出　　处：《模式识别与人工智能》1998年第1期75-81,共7页Pattern Recognition and Artificial Intelligence

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.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.

关 键 词：神经网络 预测模型 生产过程 炼铁 

分 类 号：TF5[冶金工程—钢铁冶金] TP18[自动化与计算机技术—控制理论与控制工程]