论文部分内容阅读
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。
Aiming at the problem that the supervised learning neural network represented by BP algorithm can not be asymptotically calculated in the direct multi-step prediction, a three-layer neural network model with simple feedback recursion is established. The neural network model combined with the time difference method is proposed in blast furnace hot metal Strategies for the prediction of silicon content. Combined with the real-time data collected in the field, experiments were carried out and the hit rate was higher than that predicted by the ARMAX model.