Real-time Performance Of Distributed Model Predictive Control Algorithm

Compared with traditional PID controller,Model Predictive Control(abbreviated as MPC)shows its huge advantages in processing systems with complex constraints and is widely used in the control of large-scale process industrial processes.MPC is a control method based on online optimization,For large-scale systems with strong real-time requirements,due to the large scale of optimization problems that need to be solved online,it is difficult for traditional centralized MPC to ensure that the optimized solution is converged within the specified control range which limit its scope of application.DMPC reduces the complexity of online optimization solutions by decomposing the original system into several coupled subsystems and has become one of the effective means to solve large-scale system control problems at this stage.In this paper,for linear time-invariant systems,by considering the output coupling constraints between subsystems,the centralized MPC optimization problem is transformed into a representative DMPC optimization problem.For the DMPC optimization problem,several typical DMPC optimization algorithms are discussed,the implementation steps of these algorithms are also given.The distributed ADMM algorithm combines the parallelism of the dual method and the convergence of the multiplier method,the algorithm can achieve convergence for all parameter values under certain conditions,which makes it widely used in the field of distributed optimization.This paper mainly proposes an improved distributed model predictive control algorithm for the selection of the unique parameter ? in the algorithm and analyzes the theoretical calculation of the involved algorithm: First,the original distributed ADMM algorithm is combined with the over-scaling method,and the original distributed ADMM algorithm is converted accordingly;Then,combined with the converted algorithm,the optimal step size selection principle suitable for the constructed DMPC problem is derived,and use this principle to offline compute the warm start parameters of the algorithm step size;Then,in order to further serve as the convergence speed of the algorithm,a variable penalty method is adopted in the iterative process.Finally,the simulation experiments compare the performance of the algorithm,and verify the effectiveness of the improved algorithm.