Abstract: For the controlled object with the complex and strong nonlinearity in the industrial process, a nonlinear iterative predictive control based on RBF neural network is proposed. This algorithm adopts the RBF neural network to approximate the nonlinear system, which is used as the predictive model. Meanwhile, in order to avoid missing some information of the system in each sampling time with respect to linearization, the internal predictive output along the future trajectory is expanded using the methods of Taylor series expansion and the internal iteration. Therefore, the solution of the complex nonlinear optimization is transformed into an easy quadratic programming and it can overcome the difficulty of online real-time computation of the nonlinear equation. Finally, the predictive control law is directly derived. The simulation comparison results for the CSTR process show that this algorithm has a good ability of tracking and disturbance rejection.

Key words:  , RBF neural network, predictive control, iterative, CSTR

摘要： 针对工业过程中具有复杂、强非线性的被控对象，提出一种基于RBF神经网络的非线性迭代预测控制算法。该算法采用RBF神经网络建立非线性系统过程模型，将该模型作为预测模型。同时为了避免对每个采样时刻进行线性化时会丢失系统的一些信息，因此采用多元泰勒展开和内部迭代方法，将迭代输出的预测值沿着输入轨迹展开，从而将求解复杂的非线性优化问题转化为求解简单的二次规划问题，解决了在线实时求解控制律时非线性方程的困难，最终直接递推出预测控制律的解析式。CSTR过程的仿真对比结果表明了该算法具有很强的跟踪和抗干扰能力。

关键词: RBF神经网络, 预测控制, 迭代, CSTR