A general-purpose multivariable controller using a singular value decomposition structure

A relatively new control technique, long used in the numerical analysis field, is singular value decomposition (SVD). Given the process gain matrix, the decomposition can be performed, resulting in the matrices U, Σ, and V. The U and V matrices have special properties which lend to implementation in the SVD controller. Development of this controller is briefly outlined, and studies into the practical problems of application are presented. In order to test the controller on a realistic system, a detailed nonlinear first-principles simulation of a coal-fired power plant is developed. The performance of the SVD controller is compared with that of a conventional multivariable single input-single output controller, and with another new technique, the multivariable tuning regulator. The SVD controller is found to perform quite well and, for the example of the coal-fired power plant, to be a robust algorithm.