Controle preditivo por subespaços como um serviço em nuvem.

Abstract

This work proposes an algorithm for subspace predictive controllers for networked systems, which can be extended to cloud control systems. O Subspace predictive control is a data-based control methodology that combines the advantages of model-based predictive control, such as robustness, capacity of dealing with restrictions and simplicity in the treatment of multivariable systems, but eliminating the need for a process model, which is an expensive step. The control predictive by subspaces makes use of experimental data from the process to build the its control law. To this end, the control algorithm uses projections from the identification by subspaces, which only require QR decompositions of matrices Hankel of input and output data. The proposed network control scheme consists of in a predictive controller to generate a sequence of future control signals and a network delay compensator on the actuator side. The architecture is all based on data, which minimizes identification costs and enhances the practical use of the controller in network. A new tuning method for subspace predictive controllers is proposed, employing multi-objective optimization to obtain tuning parameters that meet requirements expressed as desired time constants. Linked topics are presented the identification by open-loop subspaces and the use of this data to construct the prediction matrices. Aspects related to controller tuning and performance predictions are discussed, with the application of the tuning method developed to models generic univariables and a simplified model of a heavy oil fractionator. By Ąm, the design of a predictive controller for network subspaces is presented which is applied in a first-order univariable and multivariable models through simulations in the Simulink® environment.