基于SVM和RBF神经网络的CFB NOx生成预测模型  

NOx Generation Prediction Model of Coal-Fired Boiler Based on SVM and RBF Neural Network

作　　者：余廷芳[1] 张浩杰 YU Ting-fang;ZHANG Hao-jie(Nanchang University,Nanchang Jiangxi 330031,China)

机构地区：[1]南昌大学,江西南昌330031

出　　处：《计算机仿真》2020年第9期209-213,316,共6页Computer Simulation

基　　金：国家自然科学基金项目(6126048);江西省重点研发计划项目(2017ACG70012)。

摘　　要：针对传统燃煤锅炉NOx生成预测方法存在的预测精度低、计算复杂、泛化能力差等问题,基于某电站燃煤锅炉热态运行数据样本,分别采用RBF神经网络和支持向量机(SVM)对某燃煤锅炉NOx生成特性进行建模,在RBF神经网络模型中采用交替梯度算法对其权值及函数中心与标准偏差值进行优化;而对SVM预测模型,通过网格寻优方法对核函数及参数c和g进行选优。仿真结果表明,改进后的RBF模型和参数选优后的SVM模型的最大误差和平均误差都有了明显的降低,且计算时间快、泛化能力强。通过对两种模型的仿真和预测结果进行量化对比分析,得出SVM具有比RBF更强的预测能力及泛化能力的结论。利用建立的SVM预测模型,在设计煤种和额定负荷下,对锅炉各磨的组合方式及二次风配风方式对锅炉NOx生成特性的影响进行了仿真分析,推荐的运行方案如下:ABCDE磨组合方式及正宝塔配风运行方式下生成的NOx浓度最低。Aiming at the problems of low prediction accuracy,complicated calculation and poor generalization a⁃bility of the traditional coal-fired boiler NOx production prediction method,based on the thermal operation data sam⁃ples of a coal-fired boiler of a power station,RBF neural network and support vector machine(SVM)were adopted respectively.The NOx generation characteristics of a coal-fired boiler were modeled.In the RBF neural network mod⁃el,the weight and function center and standard deviation values were optimized by using the alternating gradient algo⁃rithm.For the SVM prediction model,the kernel function was obtained based on the grid optimization method.And the parameters c and g were selected.The simulation results show that the maximum error and average error of the im⁃proved RBF model and the parameter-selected SVM model are significantly reduced,and the calculation time is fast and the generalization ability is strong.Through the quantitative comparison and analysis of the simulation and predic⁃tion results of the two models,it is concluded that the SVM has stronger prediction ability and generalization ability than RBF.Based on the established SVM prediction model,the effects of the combination of boilers and the secondary air distribution on the NOx generation characteristics of the boiler were simulated under the design of coal type and ra⁃ted load.The recommended operation scheme is as follows:NOx concentration,generated under the ABCDE grinding combined mode and the Zhengbao tower air distribution mode,is the lowest.

关 键 词：燃煤锅炉 氮氧化物生成 径向基神经网络 支持向量机 磨煤机组合 配风方式 

分 类 号：TK223.23[动力工程及工程热物理—动力机械及工程]