钢铁企业高炉煤气发生量预测建模及应用  

Output Forecasting Model of Blast Furnace Gas in Iron & Steel Works and Its Application

作　　者：李红娟[1,2] 王建军[2] 王华[2] 孟华[2] 

机构地区：[1]昆明理工大学质量发展研究院,昆明650093 [2]昆明理工大学冶金节能减排教育部工程研究中心,昆明650093

出　　处：《系统仿真学报》2014年第6期1308-1314,共7页Journal of System Simulation

基　　金：国家自然科学基金(51066002/E060701);NSFC-云南联合基金(U0937604)

摘　　要：针对钢铁企业高炉煤气发生量的机理模型难以对其进行预测的问题,建立了基于Elman神经网络和最小二乘支持向量机相结合的预测模型。预测前利用概率神经网络对其进行分类,并对分类后的数据进行HP滤波处理,得到趋势序列和波动序列分别预测;预测后引入马尔科夫链的状态转移矩阵,对预测残差进行修正。组合建立的PNN-HP-Elman-LSSVM-MC分类预测模型,减少了训练时间,同时也提高了预测精度。根据企业实际数据应用模型,结果表明,所建模型不同工况分类准确,预测效果良好,为合理调度副产煤气提供操作依据。Aiming at the prediction for blast furnace gas output in iron and steel enterprises, which is very difficult to be modeled using the mechanism modeling, a blast furnace gas output prediction combined with the Elman neural network and the least squares support vector machine was established. Before forecast, probabilistic neural network was applied to the data classification, carrying on filter processing to the classification data, and getting the trend and volatility sequences, respectively; after forecast, the state transfer matrix of the Markov chains was introduced to adjust the residual. Assemblying the created PNN-HP-Elman-LSSVM-MC prediction model for classification, the training time was reduced, and also the precision was improved. The simulation results using the practical blast furnace gas output data in a certain iron and steel enterprise show that the predictive effect of the function is so excellent and of high classification accuracy under different conditions, and provides the reasonable scheduling byproduct gas with some operating proposals.

关 键 词：概率神经网络 HP滤波 ELMAN神经网络 最小二乘支持向量机 马尔科夫链 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]