基于改进遗传算法优化BP网络的轧制力预测研究  

Rolling Force Prediction Based on BP Network Optimized by An Improved Genetic Algorithm

作　　者：杨景明[1,2] 顾佳琪[1] 闫晓莹[1] 车海军[1,2] 

机构地区：[1]燕山大学工业计算机控制工程河北省重点实验室国家,河北秦皇岛066004 [2]国家冷轧板带装备及工艺工程技术研究中心,河北秦皇岛066004

出　　处：《矿冶工程》2015年第1期111-115,共5页Mining and Metallurgical Engineering

基　　金：河北省科技支撑计划项目(13211817);河北省高等学校创新团队领军人才培训计划项目(LJRC013)资助

摘　　要：为提高铝热连轧轧制力预报精度,满足现场生产需求,采用改进遗传算法优化神经网络建立铝热连轧轧制力的智能模型。以河南某1+4铝热连轧厂连轧实测数据作为实验样本,在遗传算法的初始化和变异机制中引入混沌序列,同时选择最优保存机制、动态调整交叉率和变异率等方法,提出了改进的遗传算法,并将其与改进的BP算法相结合,对多层前馈神经网络权值阈值进行优化,避免学习中陷入局部最小,使模型最终具有了良好的收敛性和适应性。网络预测结果与实测数据的相对误差基本在10%以内,该预测精度明显优于传统数学模型,实现了铝热连轧轧制力的高精度预测。In order to improve the prediction accuracy of rolling force of aluminum hot tandem rolling and meet the field production requirements,an improved genetic algorithm was adopted to optimize the neural network,so as to establish an intelligent model for the rolling force of aluminum hot tandem rolling. The tandem rolling test data of 1 ＋ 4 aluminum hot rolling mill in Henan was used as the experiment specimen by introducing chaotic sequence into the initialization and mutation mechanism,by choosing the best preservation mechanism and dynamically adjusting the crossover and mutation rate,an improved genetic algorithm was proposed,which was combined with an improved BP algorithm to avoid local minimum in learning by optimizing the weight and threshold of multilayer feedforward neural network,so as to make the final model of good convergence and adaptability. The relative error between the network predictions and measured data was basically within 10%. It is concluded that the prediction accuracy is significantly better than that of the traditional mathematical models,resulting in a high accuracy prediction of the rolling force of aluminum hot tandem rolling.

关 键 词：实数编码 神经网络 轧制力预报 遗传算法 混沌序列 

分 类 号：TG339[金属学及工艺—金属压力加工]