基于神经网络逆模型的广义预测控制及应用  

Generalized Predictive Control Based on Neural Network Inverse Model and Its Application

作　　者：于蒙 邹志云 朱文超 孟磊 YU Meng;ZOU Zhi-yun;ZHU Wen-chao;MENG Lei(Institute of Chemical Defense,Military Academy of Sciences,Beijing 102205,China)

机构地区：[1]军事科学院防化研究院,北京102205

出　　处：《控制工程》2021年第9期1747-1753,共7页Control Engineering of China

摘　　要：针对pH过程Hammerstein辨识模型,采用前馈神经网络辨识的逆模型控制策略。将辨识后的逆模型引入控制系统中,补偿Hammerstein模型的静态非线性,使强非线性pH过程呈伪线性特征,非线性的控制策略设计转变为线性控制策略设计。引入DE-LM算法,提高了逆辨识精度。根据伪线性结构,采用基于CARIMA模型的GPC实现控制。控制仿真实验结果表明,所建立的控制策略优于传统的PID控制算法,具有良好的设定值跟踪性能和抗干扰控制响应。For the Hammerstein identification model of pH process, the inverse model control strategy of feedforward neural network identification is adopted. The identified inverse model is introduced into the control system to compensate for the static nonlinearity of the Hammerstein model, making the nonlinear pH process pseudo-linear, and nonlinear control strategy design is transformed into linear control strategy design. The DE-LM algorithm is introduced to improve the accuracy of inverse identification. According to the pseudo-linear structure, the control is implemented by GPC based on CARIMA model. The simulation results show that the designed control strategy is superior to the traditional PID control algorithm, and has good set value tracking performance and anti-interference control response.

关 键 词：PH中和过程 HAMMERSTEIN模型 DE-LM算法 CARIMA模型 GPC 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]