欢迎访问过程工程学报, 今天是

›› 2013, Vol. 13 ›› Issue (3): 451-457.

• 系统与集成 • 上一篇    下一篇

建立PNN-HP-ENN-LSSVM模型预测钢铁企业高炉煤气发生量

李红娟 王建军 王华 孟华   

  1. 昆明理工大学 冶金节能减排教育部工程研究中心 安徽工业大学冶金与资源学院 昆明理工大学冶金与能源工程学院 昆明理工大学冶金与能源工程学院
  • 收稿日期:2013-03-18 修回日期:2013-04-07 出版日期:2013-06-20 发布日期:2013-06-20
  • 通讯作者: 李红娟

Prediction of Blast Furnace Gas Output in a Steel Complex Based on PNN-HP-ENN-LSSVM Model

LI Hong-juan WANG Jian-jun WANG Hua MENG Hua   

  1. Kunming University of Science and Technology School of Metallurgy and Resource，Anhui University of Technology Faculty of Metallurgical and Energy Engineering, Kunming University of Science and Technology Faculty of Metallurgical and Energy Engineering, Kunming University of Science and Technology
  • Received:2013-03-18 Revised:2013-04-07 Online:2013-06-20 Published:2013-06-20
  • Contact: LI Hong-juan

PDF

被引次数

25

摘要： 针对钢铁企业高炉煤气发生量机理模型难以对发生量进行精确预测的问题，通过分析高炉煤气发生量特点，按不同工况利用概率神经网络(PNN)对高炉煤气发生量进行分类，依据分类结果并结合HP滤波、Elman神经网络(ENN)、最小二乘支持向量机(LSSVM)各自的性质，建立了PNN-HP-ENN-LSSVM模型，对高炉煤气的发生量进行分类预测，并用企业实际数据验证. 结果表明，随机抽取多组测试结果中的2组，1#高炉80个点、2#高炉60个点的分类准确率分别为95%和93%，模型预测平均相对误差分别为1.0%和1.1%，适合高炉煤气发生量预测. Wilcoxon符号秩检验也验证了所提建模方法的有效性.

关键词: 高炉煤气发生量, 概率神经网络, HP滤波, Elman神经网络, 最小二乘支持向量机, PNN-HP-ENN-LSSVM模型

Abstract: Aimed at the difficult problem of accurate prediction on blast furnace gas output in an integrated iron and steel works with mechanism models available, by analyzing the gas output using probabilistic neural network (PNN) for classification according to various conditions and characteristics of probabilistic neural network, HP filter, Elman neural network (ENN) and least squares support vector machine (LSSVM), a PNN-HP-ENN-LSSVM model was established. The simulation results using the practical gas consumption data in an iron and steel complex showed that for 80 sites in 1# blast furnace, 60 sites in 2# blast furnace, the classification accuracies of 95%, and 93% were tested respectively. Then a forecasting model was founded based on the classification results to predict gas output, the average relative errors of 1.0% and 1.1% were obtained, the PNN-HP-ENN-LSSVM model was more suitable for blast furnace gas output prediction than other methods. And the Wilcoxon symbol rank test also proved the validity of the combined classification method.

Key words: blast furnace gas output, probabilistic neural network, HP filter, Elman neural network, least squares support vector machine, PNN-HP-ENN-LSSVM model

中图分类号: