基于变量选择的电站燃煤锅炉NO_(x)排放浓度预估  

Estimation of NO_(x)Emission Concentration from Coal-fired Boilers of Power Stations Based on Variable Selection

作　　者：王珑宪 赵文杰[1] WANG Long-xian;ZHAO Wen-jie(School of Control and Computer Engineering,North China Electric Power University,Baoding,Hebei 071003,China)

机构地区：[1]华北电力大学控制与计算机工程学院,河北保定071003

出　　处：《计量学报》2023年第10期1590-1596,共7页Acta Metrologica Sinica

摘　　要：针对电站燃煤锅炉NOx排放浓度存在测量迟延的情况,提出了基于互信息和长短期记忆神经网络相结合的电站燃煤锅炉NO_(x)排放浓度预测模型。首先,利用互信息计算出候选输入变量与输出变量NO_(x)浓度之间的延迟时间,并引入最大相关最小冗余算法,筛选出最优特征子集,将最优特征子集作为LSTM模型的输入,建立了锅炉NO_(x)排放浓度预测模型。仿真结果表明,所建模型的测试集均方根误差为4.626 mg/m^(3),平均绝对误差为3.836 mg/m^(3),与未经变量选择和未考虑时延的LSTM模型相比,预测精度显著提高。To study the measurement delay of NO_(x)emission concentration of coal-fired boilers in power stations,a prediction model of NO_(x)emission concentration of coal-fired boilers in power stations based on mutual information(MI)and long short-term memory neural network(LSTM)was proposed.Firstly,the delay time between the candidate input variable and the output variable NO_(x)concentration was calculated using mutual information,and the MRMR algorithm was introduced to screen out the optimal feature subset,and the optimal feature subset was used as the input of LSTM model.The prediction model of NO_(x)emission concentration of boiler is established.The simulation results show that the root mean square error(RMSE)and mean absolute error(MAE)of the proposed model are 4.626 mg/m 3 and 3.836 mg/m 3,respectively.Compared with the LSTM model without variable selection and delay consideration,the prediction accuracy is significantly improved.

关 键 词：计量学 NOx排放浓度 燃煤锅炉 变量选择 互信息 长短期记忆网络 

分 类 号：TB99[一般工业技术—计量学] TB973[机械工程—测试计量技术及仪器]