基于神经网络模型的时间序列预测算法及其应用

提出了一种神经网络模型的时间序列直接多步预测算法。网络的学习采用具有遗忘因子的ＢＰ算法与时差方法相结合的混合算法，解决了经典ＢＰ算法在直接多步预测中不能渐进计算的问题，同时网络具备一定的结构学习能力。采用该算法对现场采集的高炉铁水含硅量时间序列数据进行预报实验，表明本文提出的直接多步预测方法是可行的。

A PREDICTIVE ALGORITHM BASED ON NEURAL NETWORK MODEL FOR TIMES SERIES AND ITS APPLICATION

This paper applies a hybrid learning algorithm based on neural network model to predict time series with several steps in advance. The proposed algorithm combines time difference method with BP algorithm with forgetting. It helps to solve the computing problem incrementally in traditional BP algorithm on multi-step predicting and has the ability of structural learning. The predictions of the silicon content time series dota of the hot metal in blast furnace show that the method proposed here is feasible.