基于神经网络优化的鲁棒模型预测控制

Robust Model Predictive Control Based on Neurodynamical Optimization

在工业过程控制中,由于存在模型失配、不确定性和外部扰动等因素的影响,容易造成系统不稳定,影响控制效果。为提高控制系统鲁棒性,提出采用投影神经网络优化鲁棒模型预测控制问题。将鲁棒模型预测控制问题转化成min-max优化问题进行处理,并采用投影神经网络模型进行在线求解。该神经网络动态优化方法充分发挥了神经网络并行和分布式处理的优点,且简单结构易于电路实现,并具有全局收敛特性。通过多相流量计仿真实验,表明该算法的有效性和正确性。

Due to the presence of the model mismatch,uncertainty and external disturbance may cause the control system instability and affect the control effect. To improve the control system robustness,this paper proposes a projection neural network to solve the robust model predictive control problem. Firstly,a robust model predictive control problem is converted into a min-max optimization problem,and a projection neural network model is presented to solve it online. This neural network dynamic optimization method exerts the advantages of the neural network with parallel and distributes in solving the optimization problems. It is easy to realize using circuit and has a simple structure and the global convergence property. The GLCC multiphase flow meter simulation experiment results shows the effectiveness and the correctness of the proposed approach.