摘要
针对一类工业控制系统中存在的非线性、大时滞等情况,提出一种基于双阶段神经网络的改进隐式广义预测控制方法。首先,设计了一种基于快速回归算法和蝙蝠算法的双阶段神经网络模型,用于对非线性时滞系统进行建模,避免非线性系统下的模型失配问题;其次,采用比例积分(proportional integration, PI)结构优化广义预测控制目标函数设计,提高隐式广义预测控制性能;同时,改进控制增量选取策略,利用所预测的未来控制增量修正当前时刻控制增量;最后,将所设计的预测模型和预测控制方法应用于一个数值案例和锅炉燃烧系统,验证了所提控制策略的有效性。
In view of the nonlinear and large time delay existing in a class of industrial control systems,an improved implicit generalized predictive control method based on two-stage neural network is proposed and applied to the combustion system of gas boiler.Firstly,a two-stage neural network model establishment method based on fast regression and bat algorithm is designed to model the non-linear time-delay system to avoid model mismatch in the non-linear system.Then,in order to improve the control performance of implicit generalized predictive control,the objective function of the generalized predictive control is optimized by combining the proportional integration(PI)idea and adopting the proportional integration structure.At the same time,the control increment selection strategy of generalized predictive control is improved,and the control increment at the current time is corrected by using the predicted future time control increment to optimize the control effect.Finally,a class of nonlinear time-delay systems and boiler combustion systems are tracked and controlled by simulation.The experimental results show that the designed predictive model and control method are superior and practical.
作者
周媛奉
陈宽文
胡婷婷
刘朋远
丁海丽
梁飞
王一凡
张腾飞
ZHOU Yuanfeng;CHEN Kuanwen;HU Tingting;LIU Pengyuan;DING Haili;LIANG Fei;WANG Yifan;ZHANG Tengfei(Metrology Center,State Grid Ningxia Marketing Service Center,Yinchuan 750002,China;College of Automation&College of Artificial Intelligence,Nanjing University of Posts and Telecommunications,Nanjing 210023,China;Jiangsu Yibang Power Technology Co.,Ltd.,Nanjing 210001,China)
出处
《控制工程》
CSCD
北大核心
2024年第3期567-576,共10页
Control Engineering of China
基金
国家自然科学基金资助项目(62073173,61833011)。
关键词
非线性系统
时滞
双阶段神经网络
PI控制
广义预测控制
Nonlinear system
time delay
two-stage neural network
PI control
generalized predictive control
