为了简化约束存在时多变量广义预测控制算法的设计与实现,依据对角CARIMA模型的结构特点,将多输入多输出对象的参数辨识和模型预报问题转化为一系列多输入单输出子对象的参数辨识和模型预报问题.推导了输入输出的约束形式及优化求解过程,简化了多变量对象的参数辨识、模型预报、目标函数和约束条件系数矩阵的计算.在由DCS控制的非线性液位装置上的对比实验结果表明了该方法的有效性. In order to simplify the design and implementation of multivariable generalized predictive control algorithm when constraints exist, according to the structural features of diagonal CARIMA model, the parameters identification and model forecasting problem of multiple input and multiple output objects are transformed into a series of parameters of multiple input and single output subobjects Identification and model forecasting problems.It deduces the constraint form of I / O and the optimization solving process, simplifies the parameter identification, model prediction, objective function and constraint coefficient matrix calculation of multivariable object.In the nonlinear liquid level control device controlled by DCS The experimental results show that the method is effective.