For input and input increment constraints multivariable systems

a weight-varying based multivariable generalized predictive control of diagonal controlled auto-regressive integrated moving average (CARIMA) model was proposed to improve control performance. Because the matrices C and A of a diagonal CARIMA model are chosen to be diagonal

the prediction problem of a multi-input and multi-output (MIMO) process was transformed into generating a set of optimal predictions for a series of multi-input single-output processes

which simplified the predictive controllers and weakened the coupling of outputs in some measure. According to the difference of the model prediction value and the reference trajectory

the weight coefficients of different output tracking errors in cost function were real-time adjusted to further reduce the coupling disturbance among the outputs. The basic idea of weight coefficients adjusting was that when one output prediction was deviating from its reference trajectory at some sampling time point along the predictive horizon

the weight coefficients of all other outputs' tracing errors in the cost function increased according to the square tracing error of that output at that time point

which increased other loops' input increments at the next sampling point to eliminate the possible disturbance caused by that output deviating. Simultaneously

amplitude constraints of inputs and input increments expressions were analyzed. The validity and superiority were demonstrated by comparing simulation results for a Shell heavy oil fractionators control using common multivariable generalized predictive control (MGPC) and disturbance rejection multivariable generalized predictive control of diagonal CARIMA model proposed in the paper.