面向超临界CFB机组炉膛温度预测的仿真实验

Simulation Experiment on Furnace Temperature Prediction for Supercritical CFB Units

为开展循环流化床(CFB)机组炉膛温度预测,采用皮尔逊相关系数(PCC)法筛选输入变量,利用误差反向传播(BP)神经网络理论建立炉膛温度模型,对1台350 MW超临界CFB机组炉膛温度预测进行实验仿真研究。依据误差BP神经网络理论对机组运行数据进行处理,并利用PCC法将28个变量化简为5个;借助误差BP神经网络理论建立炉膛温度预测模型,对比分析3种不同实验仿真模型结构的预测输出、绝对误差与均方根误差。结果表明:经过PCC法化简,模型输入数量减少,BP网络温度预测效果好。该实验仿真方法有效,可用于培养学生的创新实践素养,提高学生解决工程问题的能力。

In order to study the prediction of furnace temperature for CFB units,the PCC method was used to screen input variables,and a furnace temperature model was established using BP neural network theory.Simulation experiment was conducted on the prediction of furnace temperature for a 350 MW supercritical CFB unit.We introduced the overview of CFB supercritical variable pressure operation DC boiler,the basic theory of PCC method and BP neural network,processed the operating data of the unit and simplify 28 variables into 5 variables using the PCC method.A furnace temperature prediction model was established using BP neural network theory,and the prediction output,absolute error,and root mean square error of three different simulation experimental model structures were compared and analyzed.The results show that after simplification by the PCC method,the number of model inputs is reduced,and the temperature prediction effect of the BP network is good.This simulation experimental method is effective and conducive to cultivating students’innovative practical literacy and improving their ability to solve practical engineering problems.