东北大学学报:自然科学版 ›› 2017, Vol. 38 ›› Issue (6): 778-782.DOI: 10.12068/j.issn.1005-3026.2017.06.004

• 信息与控制 • 上一篇    下一篇

神经网络预测控制在SCR烟气脱硝系统中应用

孟范伟1, 徐博2, 吕晓永1, 刘胤圻1   

  1. (1. 东北大学秦皇岛分校 控制工程学院, 河北 秦皇岛066004; 2. 吉林省电力科学研究院有限公司, 吉林 长春130021)
  • 收稿日期:2016-10-09 修回日期:2016-10-09 出版日期:2017-06-15 发布日期:2017-06-11
  • 通讯作者: 孟范伟
  • 作者简介:孟范伟(1981-),男,黑龙江青冈人,东北大学秦皇岛分校讲师,博士.
  • 基金资助:
    河北省高等学校科学技术研究项目(ZD2016203); 国网吉林省电力有限公司电力科学研究院科技项目.

Application of Neural Network Predictive Control in SCR Flue Gas Denitration System

MENG Fan-wei1, XU Bo2, LYU Xiao-yong 1, LIU Yin-qi1   

  1. 1. School of Control Engineering, Northeastern University at Qinhuangdao, Qinhuangdao 066004, China;2. Jilin Electric Power Research Institute Co., Ltd., Changchun 130021, China.
  • Received:2016-10-09 Revised:2016-10-09 Online:2017-06-15 Published:2017-06-11
  • Contact: LYU Xiao-yong
  • About author:-
  • Supported by:
    -

摘要: 以自某热电厂350MW燃煤机组的选择性催化还原(SCR)反应系统所采集的数据为依托,使用神经网络预测控制方法,研究电厂尾气中氮氧化物排放的预测及控制问题.利用神经网络的方法进行模型辨识,利用预测控制的思想对喷氨量进行控制,既可使尾气达到限排标准,亦能减少用氨量,提升经济效益的同时减少氨逃逸.采用最速梯度方法进行控制器的优化,并通过性能函数来约束控制量,达到预期输出.最后将仿真结果与现场所测数据进行对比,结果表明神经网络预测控制方案可以较准确地预测出未来有限时刻所需的喷氨量.

关键词: 选择性催化还原, 神经网络, 预测控制, 非线性自回归算法, 模型辨识

Abstract: Based on the data collected from selective catalytic reduction (SCR) reaction system of a 350MW coal-fired unit in a thermal power plant, neural network predictive control method was used to study the prediction and control of nitrogen oxides emission in power plant tail gas. Firstly, the non-linear model of SCR denitrification system was modeled and nonlinear autoregressive model was used to estimate the model. Then, by using the predictive control method to control the ammonia injection, the tail gas could achieve the standard of discharge limitation, and the amount of ammonia and ammonia escape could also be reduced, resulting in the enhancement of economic efficiency. The controller was optimized by the steepest gradient method and the control variable was constrained by the performance function to achieve the expected output. Finally, compared with the measured date in the field, the simulation results show that the neural network predictive control scheme can predict the amount of ammonia sprayed in the future at a finite time.

Key words: selective catalytic reduction (SCR), neural networks, predictive control, nonlinear auto regressive algorithm, model identification

中图分类号: