A constrained genetic algorithm for decentralized control system structure selection and optimization. (Q1413947)

From MaRDI portal
scientific article
Language Label Description Also known as
English
A constrained genetic algorithm for decentralized control system structure selection and optimization.
scientific article

    Statements

    A constrained genetic algorithm for decentralized control system structure selection and optimization. (English)
    0 references
    0 references
    0 references
    0 references
    17 November 2003
    0 references
    The authors propose the use of genetic algorithms as tools for synthesising robust decentralized control systems. The design problem is formulated as an optimization problem for a single trade-off parameter characterizing the appropriate bandwidth basis subject to constraints on the supremum of the structured singular value of the perturbation structure matrix [\textit{J. C. Doyle}, ``Analysis of feedback systems with structured uncertainties'', Proc. IEE-D 129, No. 6, 242--250 (1982)]. This matrix resulted from a transformation of the feedback feedforward structure into the perturbation structure defining robust performance. The pairings between control variables and process outputs are chosen among the candidates which satisfy Decentralized Integral Controllability conditions [\textit{M. Morari} and \textit{E. Zafiriou}, Robust process control, Englewood Cliffs, NJ, Prentice Hall (1989)]. Such a choice guarantees that the decentralized control system containing controllers with integral actions is stable and tolerant to actuator and sensors failures. The optimization problem is highly nonlinear, multidimensional, nonconvex and discontinuous in the parameter space. Moreover, the objective function has potentially a large number of local optima. To overcome these difficulties the authors transform the constrained optimization problem into an unconstrained one with a control mechanism recursively updating the value of the single trade-off parameters which drives the solution to meet the constraint. Then they use a multilevel genetic algorithm to solve the modified problem. At the master level the admissible pairings of the decentralized control systems are found while at the slave level the controller parameters are optimized for each pairing. The proposed approach was tested on the model of two coupled distillation columns taken from [\textit{L. Lang} and \textit{E. D. Gilles}, ``Multivariable control of two coupled distillation columns for multicomponent separation on a pilot plant scale'', IFAC Proc. of DYCORD+89 Symp., Maastricht, 53--60 (1989)].
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    0 references
    MIMO control synthesis
    0 references
    decentralized control
    0 references
    feedforward control
    0 references
    robust control design
    0 references
    constrained optimization
    0 references
    genetic algorithms
    0 references