基于LSTM模型的SCR系统喷氨量串级预测控制

常规PID对时变、时滞的选择性催化还原脱硝技术(SCR)脱硝系统控制效果不佳,难以满足环保排放要求,因此提出了一种基于长短期记忆(LSTM)神经网络滚动预测的串级预测控制策略。将LSTM网络预测输出作为下一时刻输入数据,建立能自动微调的SCR系统模型;将LSTM网络与预测控制方法相结合应用于SCR喷氨优化控制中,并在此优化控制方案基础上加入PID控制,建立喷氨量串级预测控制系统。仿真结果表明:该控制策略对于SCR系统具有调节速度快、动态控制性能好等优点,且能克服模型失配的影响。

Cascade Predictive Control of Ammonia Injection in SCR System Based on LSTM Model

For the time-varying, time-lagged selective catalyytic reduction(SCR) system, conventional PID control is less effective, it is more challenging to meet specification of the emission environment.Therefore, a string-level predictive control strategy based on rolling prediction of long and short-term memory(LSTM) neural networks is proposed.The output of the LSTM network was assembled into the input data of the next moment, then were new data used to model an SCR system that can be automatically fine-tuned.LSTM network was combined within predictive control methods and applied to SCR denitrification ammonia injection optimization control.PID control was added to this optimised control scheme to establish a cascade predictive control system for the ammonia injection quantity.The results show that the control strategy is fast for SCR system regulation, has good dynamic control performance, can overcome the influence of model mismatch.