基于粒子群神经网络的高炉炉温预测  

Blast furnace temperature prediction based on RBF neural network and genetic algorithm

作　　者：刘景艳[1] 张伟[1] Liu Jingyan;Zhang Wei.(School of Electricity ＆ Automation Engineering, Henan Polytechnic University, Jiaozuo 454000, Chin)

机构地区：[1]河南理工大学电气学院

出　　处：《电子测量技术》2018年第3期42-45,共4页Electronic Measurement Technology

基　　金：国家自然科学基金(U1504623)项目资助

摘　　要：高炉冶炼过程中通常以铁水硅含量来反映高炉炉温，由于影响铁水硅含量的参数较多，且各参数之间相互影响，传统的基函数（BP）神经网络在炉温预测中存在着收敛速度慢、易陷入局部极小值等缺点，提出了一种基于小生境粒子群算法优化的径向基函数（RBF）神经网络预测模型，将RBF神经网络和小生境粒子群算法有机地结合起来，利用小生境粒子群算法来优化RBF神经网络的隐层基函数宽度和中心，并利用优化后的RBF神经网络对炉温进行预测，建立了神经网络训练和检验样本集，对预测模型进行训练和检验。仿真结果表明该预测模型加快了网络收敛速度，改善了神经网络的泛化能力，具有稳定性好、预测精度高的特点。Hot metal silicon content is usually used to reflect the blast furnace temperature in smelting process. Because many parameters influence the hot metal silicon content, which have mutual influences between the parameters, the conventional BP neural network prediction method has the disadvantages of slow convergence speed, low accuracy and poor adaptive capacity. A rodial basis function （RBF） neural network prediction model based on niche particle swarm optimization algorithm is proposed. The RBF neural network and niche particle swarm optimization algorithm are organically combined. Niche particle swarm optimization algorithm is applied to optimize the width and center of the basis function, and the optimized RBF neural network is used to predict furnace temperature. A niche particle swarm neural network training and testing samples are established to train and check the prediction model. The results show that the adopted prediction model speeds up the network convergence rate and improves the generalization ability of RBF neural network, and it has the characteristic of good stability and high prediction precision.

关 键 词：炉温预测 铁水硅含量 神经网络 小生境粒子群算法 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TN911[自动化与计算机技术—控制科学与工程]