水泥窑优化控制建模问题一直是自动化界急于解决的难题。文章利用神经网络的非线性映射、特征抽取和容错能力来解决水泥窑优化控制建模问题,建立了基于BP网络的窑优化操作二层回路控制方案,并针对在BP网络训练中可能出现的一些问题如收敛、局部极小、隐层节点数确定和训练样本集对训练影响等问题进行了分析讨论,并采用了相应的训练策略,以保证训练后网络具有较好的记忆和归纳能力。仿真结果证明了采用这种策略进行训练的正确性。 Cement kiln optimization control modeling has always been a problem that the automation industry is eager to solve. In this paper, neural network nonlinear mapping, feature extraction and fault tolerance are used to solve the problem of cement kiln optimal control modeling. Based on BP neural network, a two-layer loop control scheme of kiln optimal operation is established. According to some possible problems in BP neural network training, The problems such as convergence, local minima, number of hidden nodes and training sample set are analyzed and discussed, and corresponding training strategies are adopted to ensure that the network after training has better memory and generalization ability. Simulation results demonstrate the correctness of training using this strategy.