摘要
采用BP神经网络方法预报热连轧精轧机组轧制力·通过训练数据预处理、利用遗传算法优化网络结构和参数、按钢种划分训练样本等方法,提高了网络的预报精度,优于传统的数学模型方法·BP神经网络与数学模型相结合的综合神经网络方法,进一步提高了轧制力的预报精度·预测结果与实测数据比较表明,相对误差基本在±7%以内。
Precision of prediction of rolling load on finisher plays the most important role in controlling strip thickness. To meet the increasing requirement on precision of strip thickness and shape in industry,it is necessary to improve the precision of the current prediction of the rolling load. The rolling load of a finisher was predicted by means of BP networks. Good precision of the prediction by the networks is achieved by the modeling work,such as pretreating training data,optimizing structure and parameters of networks via genetic algorithm,dividing training patterns according to steel grade. The network prediction is better than the traditional method by using mathematical models. It was proved that the combination of BP networks with mathematical models makes further improvement on the prediction accuracy. The relative prediction error is mostly between±7% from comparing the results of the prediction with measured data. The BP networks realize a high precision prediction of the rolling load on finisher.
出处
《东北大学学报(自然科学版)》
EI
CAS
CSCD
北大核心
1999年第3期319-321,共3页
Journal of Northeastern University(Natural Science)
基金
国家"九五"科技攻关项目
关键词
神经网络
遗传算法
数学模型
轧制力预报
热连轧
neural networks, genetic algorithm, mathematical models, prediction of rolling load, the finisher.