摘要
针对火力发电厂中燃烧系统运行工况复杂、迟延较大,导致选择性催化还原(SCR)烟气脱硝系统中入口NOx质量浓度难以准确测量的问题,提出了一种基于特征优化和径向基函数(radial basis function,RBF)神经网络的预测模型。将经过特征优化后的变量作为模型的最终输入变量,并使用天牛群优化(beetle swarm optimization,BSO)算法对神经网络超参数进行寻优,建立入口NO_(x)浓度预测模型。结果表明,经过特征优化后的变量放入模型后,其预测结果要优于原始变量:经特征优化及时延处理后的模型其S_(RMSE)减少了44.5%,R^(2)增加了2.3%,经过BSO确定后的神经网络超参数使得模型精度也得到了进一步提升。
In the process of thermal power generation,the operation condition of combustion system is complicated and the delay is large,which makes it difficult to accurately measure the inlet NO_(x) mass concentration in the selective catalytic reduction(SCR)flue gas denitration system.To solve this problem,a prediction model based on feature optimization and radial basis(RBF)neural network is proposed.Firstly,the variable after feature optimization is taken as the final input variable of the model.Secondly,the beetle swarm optimization(BSO)is used to optimize the neural network hyperparameters.Finally,a prediction model of inlet NO_(x) concentration is established.The results show that the predictive results of the optimized variables are better than those of the original variables.After feature optimization and timely delay,the S_(RMSE) of the model decreased by 44.5%,and the R^(2) increased by 2.3%.The neural network hyperparameters determined by BSO also improved the accuracy of the model.
作者
张国兴
王世朋
ZHANG Guoxing;WANG Shipeng(Guoneng Ningxia Yuanyanghu No.1 Power Generation Co.,Ltd.,Yinchuan,Ningxia 750011,China)
出处
《计量学报》
CSCD
北大核心
2024年第2期285-293,共9页
Acta Metrologica Sinica
基金
国家重点研发计划(2018YFB0604300)。
