Fast Model Predictive Control Algorithm Based On State Space Model

Model Predictive Control is an advanced Control proposed in the1970s from practical industry application. It had solved many issues which traditional PID algorithm and modern control theory couldn’t solve in the industry control field. However, the complex on-line optimization brings a heavy computation burden which extremely limited the application of MPC algorithm. Many researchers tried to improve the algorithm from the mathematic vision or tried to adjust the optimization structure to reduce the computation burden. Some of the methods are efficient, but some of them are too complicated to realize. So, the situation of reducing the computation problem of MPC in the practical application is still not ideal.In order to optimize this problem, a fast algorithm is proposed which modulate the real-time values into1-bit streams with Delta-Sigma modulator to simplify the multiplication operation. The one-bit based operations can instead the conventional multiplication operation by changing the sign of the multi-bit coefficient which could extremely decrease the compute consumption. In addition, the theory of deducing the matrix dimension when solve the QR optimizations is modified to integrate the1-bit based fast MPC algorithm. This method can greatly improve the efficiency of MPC algorithm by reducing the matrix demission and using the1-bit based operation. The control structure of fast MPC algorithm proposed in this paper is designed. The comparison demonstrated the superiority of the fast MPC algorithm in the aspect of reducing the computation problem. Furthermore, the control structure based on the demodulated control signals of fast MPC algorithm is designed to extend the application of the fast MPC algorithm proposed in this paper. At last, the importance of researching the1-bit based model identification theory is pointed out. In addition, the1-bit based fast operation is applied into the model identification theory successfully. The theoretical comparison and the practical simulink demonstrate its feasibility. The result is beneficial not only for solving the computation burden of model identification, but also the on-line application.