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 multi-layer local regression neural network is introduced and a new prediction model of sintering process is established by using this network. A new learning algorithm is proposed, which is called adaptive gradient adaptive gradient descent method, which accelerates the convergence process. In the meantime, Network, and compares its dynamic modeling ability with adaptive time-delay neural network (ATNN) .The simulation results show that this network has strong anti-noise ability and strong dynamic modeling ability , So it is suitable for the modeling, forecasting and control of complex production processes.