Structural Characteristics Analysis Of Two-layered Predictive Control

The modern large-scale industrial systems adopt the hierarchical structures, where the advanced process control, with model predictive control (MPC) as its representative, has become one of the important levels. With the development of industry technology, original standard MPC controller is replaced with an improved, two-layered predictive controller which consists of steady-steady optimization. Its ability to perform local economic optimization based on dynamic prediction so that the two-layered predictive controller can not only lower the variance of process values but also increase much extra economic performance is considered as a main advantage. Research about MPC in recent years pay more attention to the design, or performance of its dynamic control algorithm, and research about steady-steady optimization focus on feasibility analysis. However, issues about the function mechanism between steady-steady optimization and dynamic control receive little attention.In this paper, the function mechanism between steady-steady optimization and dynamic control is studied from the aspect of steady-state, steady-state analysis of stable process and integral process is performed, thus the insight about two-layered design is exposed. And performance comparison between two-layered predictive control and range control is performed, and singularity problem of two-layered predictive control is studied. The main wok and its corresponding results is listed as follows:1. Multi-variable control system can be divided into square system, fat system and thin system according to inputs’and outputs’numbers in structure form, existence of compatibility and uniqueness issues in un-square system controlling is illustrated with simulation examples, root cause of compatibility and uniqueness problem is analyzed with the help of steady-state balance, thus the theoretical meaning of steady-state optimization is shown, and solution for compatibility and uniqueness problem based on two-layered predictive control is given.2. A criterion, which can be used to determine whether set-points could be reached when controlling multivariable processes, is proposed according to the steady-state analysis of predictive control algorithm for integral process, and the effect of model-plant mismatch and disturbance on both optimization and control is discussed. The two-layered MPC algorithm is improved with the compensation factor so that the optimization and control algorithm can be used to control integral process with model-plant mismatch.3. There exist two different strategies in model predictive control (MPC), i.e., set-point control and range control, And range MPC is thought to be a better control strategy. The principles of range control and double-layered MPC are analyzed from the aspect of process steady state, differences and similarities between these two strategies both qualitatively and quantitatively are discussed, insight view that the two strategies are consistent in a certain context is brought up and proved.4. Ill-conditioned model is a common problem in model predictive control, a relative insight view that controlled outputs’move direction is a key factor about whether the ill-conditioned model will have a negative effect on controller’s performance is brought up, geometry tools and SVD in linear algebra is used to discover the reason of the pre-mentioned phenomenon, and an offline strategy is proposed to resolve the ill-conditioned model problem, simulation examples are used to verify the conclusion.