基于粒子群算法的低碳铬铁磷含量预测研究  

Investigation on Prediction of Phosphorus Content of Low Carbon Ferrochrome with Particle Swarm Optimization

作　　者：邱东[1] 刘明硕[1] 郭红涛[1] 

机构地区：[1]长春工业大学电气与电子工程学院,吉林长春130012

出　　处：《计算机技术与发展》2017年第6期142-145,共4页Computer Technology and Development

基　　金：吉林省科技发展计划项目(20120420)

摘　　要：低碳铬铁合金冶炼是一种高耦合多相位的物理化学过程,磷是其工艺过程中影响铬铁合金产品质量的主要杂质之一。为实现降低磷含量并提高铬铁合金的产品质量,以神经网络预测理论为指导,以RBF人工神经网络作为AOD炉冶炼过程预测和系统辨识的途径,基于某铁合金公司的生产样本数据,建立了磷含量的神经网络在线预测平台,将其预测的输出值与实际样本值之间的灰色关联度作为研究的目标函数,并利用改进的粒子群算法(PSO)解决了一般RBF神经网络出现局部最优的问题,使得磷含量预测误差明显减少,实现了对磷含量的优化控制。研究结果表明,所建立改进的PSO优化预测控制模型精度提高到95.4%,分散度在±0.003%之内,为改进冶炼工艺、提高铬铁合金产品质量提供了重要的预测手段。Low carbon ferrochrome smelting is a physical and chemical processing with extreme coupling and multiphase. The discretion of the phosphorus＂ impurity content is an important factor which affects the quality of ferrochrome products. In order to reduce the content of phosphorus and raise the quality of ferrochrome, based on the production sampling data from a ferroalloy company, an online prediction platform of neural network of phosphorus content is set up with the neural network prediction theory as the guidance and RBF neural net- work as the approach of AOD furnace prediction and system identification. The gray correlation between predicted ouput and actual sam- piing is selected as the objective function. At the same time, the modified Particle Swarm Optimization （PSO） has been used to solve an local optimization problem of the general neural network, appratent reduction of predictive error of phosphorus content and implementa- tion of its optimized control. The simulation results show that the optimization control model is 95.4% within the dispersion of ±0. 003% ,which has realized effectively phosphorus content optimal control,which can provide an important theoretical support for impro- ving the aualitv of low carbon ferrochromium.

关 键 词：粒子群算法 RBF神经网络 AOD炉 预测控制 

分 类 号：TF533.2[冶金工程—钢铁冶金]