基于PSO-BP神经网络的高炉煤气柜位预测模型及应用  

BFG holder forecasting model and application based on PSO-BP neural network model

作　　者：魏津瑜[1] 张玮[1] 李欣[2] 

机构地区：[1]天津理工大学管理学院,天津300384 [2]潍坊科技学院经济管理学院,山东寿光262700

出　　处：《中南大学学报（自然科学版）》2013年第S1期266-270,共5页Journal of Central South University:Science and Technology

摘　　要：以钢铁企业高炉煤气柜位作为研究对象,对高炉煤气发生量和消耗量等影响因素进行相关性分析,得到关联度较高的主成分作为输入参数。利用粒子群算法优化BP神经网络的初始权值和阈值,使优化后的BP神经网络能够更好地预测函数输出。针对粒子群算法早熟收敛、搜索精度偏低以及后期迭代效率不高的问题,提出对粒子进行变异操作的处理方法。研究结果表明:所建立的预测模型误差小,能有效解决高炉煤气柜位的预测问题,为生产企业合理制定煤气使用计划提供理论依据。Taking the iron and steel enterprise’s BFG holder as a research object,the correlation between influencing factors of the generation and consumption of BFG was analyzed for the principal components with higher degree.Particle swarm optimization was used to optimize BP neural network’s initial weights and thresholds in order to get better forecasting function output.Aiming at the problems of PSO’s premature convergence,low searching accuracy and bad late iteration efficiency,the methods of mutation operation on particles were proposed.The experiment result shows that the error of the forecasting model is smaller and the method can effectively solve BFG holder’s forecasting problem for providing the production with the theory basis on a reasonable gas using plan.

关 键 词：高炉煤气柜位 相关性分析 预测模型 粒子群优化 BP神经网络 变异操作 

分 类 号：TF573[冶金工程—钢铁冶金] TP183[自动化与计算机技术—控制理论与控制工程]