【摘要】 针对一类具有分数传输滞后、参数时变的典型工业过程开环参数与广义预测控制器参数之间的内在联系 ,通过在目标函数中引入开环增益 ,并利用BP网络的非线性映射能力得到了自适应广义预测控制的一种直接算法。在该算法中 ,先用一个辨识器辨识过程开环参数 ,然后由一个已训练好的BP网络根据辨识结果和加权因子的值直接计算出控制器的参数 ,得到控制律。该方法不依赖于过程的精确模型 ,极大地简化了在线计算负担。同时 ,把区域控制的思想引入到预测控制之中 ,给出了两种区域预测控制方案。在一个二元精馏塔模型上与常规广义预测控制方案的对比仿真结果验证了文中所示方法的可行性更多还原

【Abstract】 After discussing the intrinsic relationship between the open-loop parameters in a class of typical industrial processes with fractional delay and time-variable properties and the parameters in the generalized predictive controller, a generalized adaptive predictive direct algorithm based on neural networks’ mapping ability is presented by adding the open-loop system gain to the cost function. In this algorithm, an identifier is employed to estimate the parameters of the open-loop system. The parameters of the controller is directly calculated by using the identification results and the value of the control weight factor from a trained BP neural network. And then the control law is obtained. The method developed in this paper does not only depend on the exact model of the controlled plant, but also can substantially reduce the computation load on-line. Meanwhile, two kinds of zone predictive control schemes are presented as well by introducing the zone control to the predictive control. The simulation comparison with conventional GPC on a binary distillation column model demonstrates the feasibility of this algorithm.更多还原