Research Of Model Predictive Control Algorithm Based On Learning

Model predictive control is an effective way to deal with constraints and achieve excellent tracking performance. However, if there exists model errors, unreasonable parameters in the algorithm or the control law, it may affect the accuracy and the speed of the control. Learning control mainly aims at the engineering objects with cycle characteristics. It can utilize the reference trajectory or the periodic characters of the disturbance to make use of previous information to design a new control action, thus the control performance and the tracking accuracy can be improved. When dealing with the plants with cycle characteristics, if predictive control and learning control are combined, the advantages of these two algorithms can be reached. It can enhance the robustness of the system and achieve fast tracking with high precision. Based on these ideas, this article will analyze and study the problems appeared in the model predictive control algorithm based on learning control. The main works established are as follows.1) Multi-rate model predictive control to reject periodic disturbance with reduced ripple. Firstly, analyze the ripple situation in the repetitive control and proposed the multi-rate model to improve this phenomenon; secondly an augmented model of the plant is obtained containing disturbances with double sampling rate; finally, solve the new system using MPC based on the state space and implement the receding horizon control. A case study on a continuous system has demonstrated that with double sampling rate, the control is more effective than using haploid sampling rate.2) Repetitive predictive control for systems subject to periodic disturbance with Markov jump uncertainty. Firstly, a Markov jump model of the disturbance is introduced and an augmented model of the plant is obtained; secondly, solve the new system using MPC based on the state space and implement the receding horizon control. The simulation has demonstrated that this method is effective in statistical sense.3) Iterative learning predictive control for urban drainage systems. Firstly, introduce the mathematical model of the urban drainage system and get the simplified model; secondl, analyze the rainfall and divide it into three modes, proposed the iterative learning predictive control algorithm; finally, use MPC to solve this optimazation and analyze the control results.