【摘要】 针对以ＢＰ算法为代表的监督学习神经网络在直接多步预测中不能渐进计算的问题，建立了一个三层简单反馈递归的神经网络模型，提出了将神经网络模型与时差方法相结合在高炉铁水硅含量预报中应用的策略。结合现场采集的实时数据进行实验，并与采用ＡＲＭＡＸ模型的预测结果相比较，具有较高的命中率。更多还原

【Abstract】 BP algorithm is the typical supervised learning algorithm. It can not be computed incrementally for real time prediction in steps ahead Therefore, a three layer simple recurrent network model is proposed and a method combining neural network model with temporal difference method is applied to predict the silicon content of the hot metal Using the collected real time data,the proposed method gives better results compared with ARMAX models更多还原