Block factorization of step response model predictive control problems. (May 2017)

• Record Type: Journal Article
• Authors: Kufoalor, D.K.M.; Frison, G.; Imsland, L.; Johansen, T.A.; Jørgensen, J.B.
• Is Part Of: Journal of process control. Volume 53(2017)
• Page Start: 1
• Page End: 14
• Publication Date: 2017-05
• Subjects: Model predictive control -- Step response models -- Block factorization -- Interior-point methods -- Riccati recursion -- Condensing -- Numerical optimization; Process control -- Periodicals; Fabrication -- Contrôle -- Périodiques; Process control; Periodicals; Electronic journals; 660.281
• DOI: 10.1016/j.jprocont.2017.02.003
• Languages: English
• ISSNs: 0959-1524

Abstract

Abstract : Highlights: A novel MPC scheme that uses step response data in a traditional manner is proposed. The MPC scheme enables the use of block factorization in efficient QP solver methods. Riccati recursion and condensing algorithms for interior-point methods are proposed. Exploitation of the step response MPC problem structure leads to high performance. High performance is achieved for an industrial problem using the HPMPC framework. Abstract: By introducing a stage-wise prediction formulation that enables the use of highly efficient quadratic programming (QP) solution methods, this paper expands the computational toolbox for solving step response MPC problems. We propose a novel MPC scheme that is able to incorporate step response data in a traditional manner and use the computationally efficient block factorization facilities in QP solution methods. In order to solve the MPC problem efficiently, both tailored Riccati recursion and condensing algorithms are proposed and embedded into an interior-point method. The proposed algorithms were implemented in the HPMPC framework, and the performance is evaluated through simulation studies. The results confirm that a computationally fast controller is achieved, compared to the traditional step response MPC scheme that relies on an explicit prediction formulation. Moreover, the tailored condensing algorithm exhibits superior performance and produces solution times comparable to that achieved when using a condensing scheme for an equivalent (but much smaller) state-space model derived from first-principles. Implementation aspects necessary for high performance on embedded platforms are discussed, and results using a programmable logic controller are presented.