神经网络应用于系统建模时要考虑两个关键问题：一是采用的神经网络类型；二是当神经网络类型确定以后，确定网络的输入向量，这两个问题是紧密关联的。相对于BP神经网络收敛太慢，具有泛化能力的缺点，DRNN网络(对角递归神经网络)能实现动态非线性映射，具有记忆功能，可以追踪模型的变化，具有更好的预测效果。首先用关联规则方法挖掘出一些与质量指标有关的工艺条件，再结合现场工人的实际经验，找到模型的输入输出，再运用DRNN对82B钢进行建模和优化，取得了很好的预测效果。
Two key problems of modeling nonlinear system using neural network are discussed. One is the type of neural network, the other is to get the input of network when the type of network is defined. Because the BP network can′t update in time and has slow rate of convergence, DRNN network can realize the nonlinear map dynamically, and can trace the change of the model, so it has better predicting result. It first excavated some technological conditions related to the quality index by association rule, then the appropriate input and output of the model could be found by combining the worker′s experience, finally the 82B steel was optimized using DRNN (Diagonal Recurrent Neural Network) network, and a better result was obtained.
参考文献
- 下载量()
- 访问量()
- 您的评分：
-
10%
-
20%
-
30%
-
40%
-
50%