链箅机预热段温度场模型的贝叶斯-BP神经网络系统辨识  

Identification on Bayes-BP neural network system for temperature field model of preheating section of chain grate

作　　者：张铭 修晓波 周峰 李伯全[1] Zhang Ming;Xiu Xiaobo;Zhou Feng;Li Boquan(School of Mechanical Engineering,Jiangsu University,Zhenjiang 212013,Jiangsu)

机构地区：[1]江苏大学机械工程学院,江苏镇江212013

出　　处：《烧结球团》2020年第5期44-48,70,共6页Sintering and Pelletizing

基　　金：国家自然科学基金资助项目(51675245);江苏大学第19批大学生科研课题立项资助项目(19A178)。

摘　　要：针对球团链箅机预热段温度场因非线性、时滞性、不确定性等特点难以通过传统的理论分析方法建立数学模型的问题,建立贝叶斯-BP神经网络,对该温度场模型进行系统辨识,对比模型预测输出值与实际系统输出值,通过仿真与实验分析该辨识模型的拟合效果。结果表明:贝叶斯-BP神经网络拟合效果较好,其线性拟合度近似为1,最终预测误差约为0.014 K,预测相对误差在5%范围内,构建的预热段温度场模型准确可靠且适用性强,可为预热段温度场均衡稳定控制提供理论指导。Aiming at the problem that the temperature field of the preheating section of the chain grate for pellet features non-linearity,time lag,uncertainty,etc.,which is difficult to establish its mathematical model through traditional theoretical analysis methods,a Bayesian-BP neural network is established to systematically identify the temperature field model,compare the predicted output value of the model with the actual system output value,and conduct simulation and experiment on the fitting effect of the identification model.The results show that the Bayesian-BP neural network has a good fitting effect,the linearity is approximate 1,the final prediction error is about 0.014 K,and the relative prediction error is within 5%.The constructed temperature field model of preheating section is accurate,reliable and applicable,which can provide theoretical guidance for the balanced and stable control of the temperature field of the preheating section.

关 键 词：链箅机 温度场模型 系统辨识 贝叶斯-BP神经网络 最终预测误差 预测相对误差 

分 类 号：TF046.6[冶金工程—冶金物理化学] TP18[自动化与计算机技术—控制理论与控制工程]