基于入口NOx质量浓度修正的脱硝系统多模型预测控制策略及其应用案例

建立不同锅炉工况下基于欧式距离聚类的总给煤量长短期记忆神经网络预测模型，对入口处的NO_x质量浓度数据进行修正，其验证集上的均方根误差为3.53 mg·m^-3。该结果优于常见的回归方法深度神经网络与随机森林回归的预测结果。以此为基础，研究了基于入口NO_x质量浓度修正的多模型预测控制(MMPC)，设计并进行了脱硝系统仿真。仿真结果表明，与原有控制条件相比，基于入口NO_x质量浓度修正的MMPC策略使脱硝系统出口NO_x质量浓度波动幅度减小了63.7%，并能满足出口质量浓度指标为40 mg·m^-3时的控制要求，实现卡边控制。现场工程应用结果表明：在高、中、低负荷正常运行的工况条件下，入口修正-MMPC策略可将出口NO_x质量浓度波动分别控制在±10.6 mg·m⁻³、±5.5 mg·m⁻³、±4.9 mg·m⁻³，以标准差来衡量波动幅度即分别减小了53.4%、74.7%、64.6%，此控制水平优于原有控制效果；在出口NO_x质量浓度易超标并出现高浓度氨逃逸的快速变负荷工况下，升、降负荷出口NO_x质量浓度波动分别控制在±6 mg·m⁻³、±5 mg·m⁻³，此控制水平仍优于原有控制效果。本研究的入口修正-MMPC控制策略可实现不同负荷、工况下的喷氨控制，减小出口NO_x波动幅度，降低后续设备低温腐蚀的风险，从而提高SNCR/SCR联合脱硝系统运行的经济性和安全性。

Multi-model predictive control of denitrification system based on modification of inlet NOx mass concentration and its application case

A composite long short-term memory neural network was trained to correct the lag of inlet NOx mass concentration, on dataset divided by the Euclidean distance of the total coal supply of the boiler under typical working conditions. The root mean squared error on the verification dataset was 3.53 mg·m−3, which was better than the prediction results of common regression methods, such as deep neural network and random forest regression. On this basis, the Multi-model Predict Control (MMPC) strategy based on the correction of the inlet NOx mass concentration was studied, and the denitration system simulation was designed and carried out. The simulation result shows that compared with previous control, the MMPC with inlet NOx mass concentration correction makes the fluctuation range of outlet NOx mass concentration reduced by 63.7%, and it can also meet the control requirements when setting outlet mass concentration at 40 mg·m−3. The results of field application indicated that under operation conditions of high, medium and low load, the MMPC with inlet NOx mass concentration correction can control the outlet NOx mass concentration fluctuation between ±10.6, ±5.5 and ±9 mg·m-3, and the fluctuation range was reduced by 53.4%, 74.7% and 64.6% respectively by standard deviation. Then, under the condition of fast variable load, it is easy to happen that emission of NOx mass concentration at the outlet exceeds the standard and ammonia escapes at high concentration. However, the fluctuation of NOx mass concentration at the outlet of load up and load down coditionis controlled respectively within ± 6 and ±5 mg·m−3 by MMPC with inlet NOx mass concentration correction. It proves that the control performance is better than the original control effect. Above all, under different loads and operation conditions, the MMPC with inlet NOx mass concentration correction strategy studied in this paper can well control the amount of ammonia injection, reduce the fluctuation of NOx mass concentration at the outlet, minimize the risk of low temperature corrosion of subsequent equipment, and consequently improve the economy and safety of SNCR/SCR coupling denitration system.