Research Of Predictive Control Based On Polyhedral Model

Model predictive control has been widely used in process control application andachieved substantial beneficial effects. However, since the large computational load caused bythe reduplicative online optimal operation and systems in practical application always containnonlinearity, constraints and time delays, the traditional model predictive control is almostimpossible to be applied into the systems with large sampling time and intraday variationstates in electron and aviation fields.Model predictive control based on the multi-parameter programming combines onlineform with off-line and compensated the disadvantages of model predictive control efficiently.In the model predictive control method, the problem of quadratic optimization of constrainedlinear time-varying system is solved using multi-parameter programming method to obtainthe convex division of the system state space and discrete-time explicit solution. Then, thecomputation load of controller can be mitigating by looking up the control rate table in anoff-line fashion. In such a framework, we are only required to determine which regimecontains the current state, making the resulting method attractable in application.The existing results of the model predictive control based on the multi-parameterprogramming for linear time-invariant system are abundant at both home and abroad.However, system is nonlinear is rarely reported. Therefore, in this paper, the neural network iscombined with the model predictive control, the main research is present below.1. Further research on the model predictive control of linear time-invariant system.Divide the region of state in an off-line fashion, and then track and control the state online.2. Research on the model predictive control of linear time-varying system. The influencecaused by the time-variant parameter matrix is eliminated using the max-min method. Themulti-parameter programming is resorted to An improvement on the parameter-decreasemethod in presence of distribution. The stability and effectiveness of the proposed method isillustrated using a simulation example in different situations.3. According to the problem that it is difficult to design controller for nonlinear system,we take the advantages of neural network in modeling the nonlinear system and obtain thelinear models of apex of the polyhedral. An inverted pendulum system is utilized to showvalidity of our method.