摘要
针对前馈神经网络模型的BP算法自身存在的易陷入局部极小和收敛速度慢等缺陷,耦合同伦理论与BI神经网络模型,建立了基于同伦BI神经网络的转炉终点预测模型分别对转炉终点钢水的碳含量及温度进行预测,并在相同构造下同单纯BI网络模型相比较.结果表明:在网络结构相同的条件下,耦合同伦算法后预测模型的精度得到显著提高,各模型命中率的平均提高量分别为8.6%,20.2%,预测误差绝对值的最大值分别下降了48.4%,44.76%;在计算效率方面,完成相同的计算迭代次数,同伦模型所需时间平均减少14%.
Aiming at the defects of the neural network itself,such as low speed converging and liable to be trapped in local minimum,a homotopic BI neural network model is developed by combining the homotopy theory and the BI neural network model to predict the end-point carbon content and temperature of molten steel in BOF steelmaking process.Comparison with the simple BI model shows that the precision of new model is significantly improved.The hit rates are increased by about 8.6% and 20.2%,and the prediction residuals have decreased 48.4% and 44.76% respectively.Also,the calculation time of the new model is 14% shorter than BI model.
出处
《材料与冶金学报》
CAS
2010年第2期92-96,共5页
Journal of Materials and Metallurgy
关键词
转炉模型
终点预测
神经网络
同伦算法
BOF model
end-point prediction
neural network
homotopy algorithm
