Research And Implementation Of Model Predictive Control Based On Fast Gradient Method

A class of online optimization algorithms based on fast gradient method can be used to solve the online real-time model predictive control problem online with its ability to predict the upper bound of computational complexity.However,due to the different MPC control problem formulations and the differences in algorithm design,It is difficult to compare the computational effort of each algorithm.As a result,it is difficult to effectively select the appropriate algorithm for a specified MPC problem.In order to compare the existing algorithms effectively,the main components of each algorithm and its applicable MPC problem formulation has been studied,and the advantages and disadvantages of each algorithm presented.For the MPC problems with general state constraints and input constraints,the computational complexity of each algorithm is given.By using a simulated spring-mass problem as the test bed,for different prediction time domains and the size of the termination error,the number of iterations and the solution time for each algorithm has been recorded.The simulation result verify the theoretical analysis result.Combining the advantages of the existing algorithms,a new improved fast gradient algorithm is proposed.This algorithm can handle the MPC control problem with general input-state constraints,and can deal with the ill-conditioned MPC problem effectively with fast convergence speed.The memory requirements and computational performance of the algorithm are analyzed and discussed.The effectiveness the algorithm is verified by examples.