辅助给水系统神经网络模型预测控制方案设计  

Design of Neural Network Model Predictive Control Scheme for Auxiliary Feedwater System

作　　者：李鸣谦 王硕 陈日罡 LI Mingqian;WANG Shuo;CHEN Rigang(China Nuclear Power Engineering Co,.Ltd,.Beijing 100840,China)

机构地区：[1]中国核电工程有限公司,北京100840

出　　处：《自动化仪表》2023年第S01期333-336,共4页Process Automation Instrumentation

摘　　要：为优化“华龙一号”核电机组辅助给水系统控制性能,对辅助给水控制方案进行了研究,设计了一种基于长短时记忆(LSTM)网络的辅助给水系统神经网络模型预测控制方案。通过仿真验证平台进行数据采集后,利用LSTM神经网络训练出蒸汽发生器液位预测器,以反应控制过程的动态特性。应用模型预测控制方案,根据蒸汽发生器液位估计值实时优化控制变量。验证结果表明,与手动控制/核电厂自动控制方案相比,模型预测控制方案可实现更快、更准确的设定点跟踪功能,同时需要的控制工作量更少,可达到有效减少对系统部件的磨损、延长系统部件使用寿命的作用。此外,所提方案可被应用于其他需要精确和有效控制的复杂工业过程。To optimize the control performance of the auxiliary feedwater system of HPR1000 nuclear power unit,the auxiliary feedwater control scheme is studied,and an auxiliary feedwater neural network model prediction control scheme based on long short⁃term memory(LSTM)network is designed.After data acquisition through the simulation and validation platform,a steam generator level predictor is trained using the LSTM neural network to respond to the dynamic characteristics of the control process.The model predictive control scheme is applied to optimize the control variables in real time based on the estimated steam generator level.The validation results show that the model predictive control scheme can achieve faster and more accurate setpoint tracking function than the manual control/nuclear power plant automatic control scheme,while requiring less control effort,which can achieve an effective reduction of wear and tear on system components and prolong the service life of system components.In addition,the proposed scheme can be applied to other complex industrial processes that require precise and efficient control.

关 键 词：核电厂 神经网络 长短时记忆 模型预测控制 辅助给水系统 智能控制 

分 类 号：TH-702[机械工程]