钢铁企业烧结工艺蒸汽产量预测模型研究  

Study on Steam Production Forecast Model for Sintering Process in Iron and Steel Enterprise

作　　者：岳有军[1] 李佳佳[1] 赵辉[1] 王红君[1] 

机构地区：[1]天津理工大学天津市复杂控制理论与应用重点实验室,天津300384

出　　处：《自动化与仪表》2016年第2期9-11,36,共4页Automation & Instrumentation

基　　金：天津市科技支撑计划项目(13ZCZDGX03800;10ZCKFGX03400);天津市自然科学基金项目(09JCZDJC23900)

摘　　要：在烧结蒸汽产量预测问题的研究中,由于受到烧结工艺运行特点的影响,使得蒸汽具有波动大、非线性的特点。而传统的经验法预测精度较低,为了提高预测精度,结合逐步回归分析方法和小波分析的BP神经网络构建蒸汽预测模型,先通过逐步回归对自变量进行筛选,以提高网络的泛化能力;然后利用小波函数取代BP网络中的激活函数,克服神经网络易陷入局部次优的缺点,提高模型的预测精度和准确度。实验结果证明,逐步回归-小波神经网络模型的预测结果优于传统BP网络模型和小波神经网络模型,表明该模型收敛速度快、预测精度高,可为蒸汽生产和实时调度提供决策依据。As affected by the sintering process operation characteristics,the prediction of the output of the steam has large fluctuation and nonlinearity. The problem of low accuracy exist in the traditional experience,in order to improve the forecasting accuracy,combine with stepwise regression analysis method and wavelet analysis of BP neural network to build steam prediction model. Firstly,the independent variables are filtered by stepwise regression to improve the generalization ability of the network;then using the wavelet function to replace the activation function of BP network, which to overcome the disadvantages of the neural network is likely to fall into the local hypo-strongpoint,improve the prediction precision and accuracy. The results of the experiment prove that stepwise regression-wavelet neural network is superior to the traditional BP network and wavelet neural network in prediction performance,which shows that the model convergence speed is fast,high prediction accuracy,which can provide decision-making basis for steam production and real-time scheduling.

关 键 词：非线性 蒸汽模型 逐步回归 小波神经网络 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]