摘要
喷氨量大小不仅影响超临界锅炉选择性催化还原(selective catalytic reduction,SCR)烟气脱硝装置的效率,过量喷氨也会导致下游空预器受热面的积灰、腐蚀和造成资源浪费、二次污染,且在变负荷时,传统PID控制方式很难实现最佳控制。通过引入混结构隐含层,改善传统RBF神经网络变工况控制时的非线性和扰动适应能力,设计了基于混结构RBF神经网络(MS-RBFNN)的喷氨流量最优控制系统,用MS-RBFNN综合学习当前主要相关状态参数,以SCR脱硝装置出口NOx排放量最小作为学习训练信号,实时并行计算出最优喷氨控制流量。实验结果表明,此优化方案相对传统PID控制,具有更好的NOx排放控制效果和变工况适应能力,同时节约了喷氨量。
Spraying ammonia flow can influence the efficiency of supercritical boiler＇s flue gas denitrification device based on selective catalytic reduction （SCR）. Excessive spraying flow can also result in ash deposit and corruption of backward heating units such as air heater, simultaneously, it causes resource waste and second pollution. Moreover, optimal traditional PID control with variational load on the flow is difficult. And in order to improve traditional radial basis function （RBF） neural network （RBFNN）＇s adaptivities of nonlinearity and disturbance during variational working condition, so, a new control scheme based on mixed structure RBFNN （MS-RBFNN） was proposed. This MS-RBFNN can synthetically study current main relative state parameters, so as to parallel calculate the optimal spraying ammonia flow by using least NOx discharge of SCR device as its training signal. Experimental results indicate, comparing with traditional PID control, this scheme＇s advantages on better NOx control effect and adaptability of variable working condition as well as little ammonia usage.
出处
《中国电机工程学报》
EI
CSCD
北大核心
2011年第5期108-113,共6页
Proceedings of the CSEE
关键词
选择性催化还原
径向基函数神经网络
混结构
最优控制
烟气脱硝
超临界锅炉
selective catalytic reduction （SCR）
radial basis function （RBF） neural network
mixed structure
optimal control
flue gas denitrification
supercritical boiler
作者简介
周洪煜（1954），男，博士，硕士生导师，研究方向为电站控制系统、节能减排和智能控制理论，quzhy@cqu．edu．cn．
张振华（1979），男，硕士研究生，研究方向为电站DCS控制系统优化和智能控制应用；
张军（1971），男，硕士，高级工程师，研究方向为清洁发电技术和神经网络控制。
