在传统BP神经网络预测模型的基础上，依据灰色理论中的灰色关联度，提出了输出变量各个影响因素的灰色关联度权值，首次建立基于灰色理论的神经网络预测模型，并依据国内某钢厂300组实际生产数据进行仿真试验。试验结果表明：误差绝对值小于5%的炉数有39炉，占总炉数的65.00%；误差绝对值小于10%的炉数共有58炉，占到96.67%。与传统BP神经网络相比，基于灰色理论的神经网络模型的预测精度提高近12.5%，说明基于灰色理论的铁水预处理终点磷含量神经网络预测模型能更精确地反映现场实际水平。
On the basis of optimizing the traditional BP neural network model, a neural network prediction model combined with grey theory was developed by laying correlative degree weights to all input factors which had effects on the output variable. And then simulation experiments of model newly established were conducted based on data from a domestic steel plant. The results show that hit rate arrives at 65.00% when the error modulus is less than 500%, and the value is 96.67% when less than 10.00%. Comparing to the traditional neural network prediction model, the accuracy almost increases by 12.50%, Thus, the prediction of end point phosphorus content fits the real perfectly, which accounts for that neural network model for terminative phosphorus content based on grey theory can reflect accurately the practice in hot metal pretreatment.
参考文献
- 下载量()
- 访问量()
- 您的评分：
-
10%
-
20%
-
30%
-
40%
-
50%