Model design and data analysis for multi-input multi-output systems

The design of multivariable control systems requires identification of the effects of individual inputs on each of the outputs. In many complex systems whose behavior is described by a large set of partial differential equations, the solution cannot be implemented in real time. We present an overview of several linear and nonlinear approximators: least squares, principle component regression, partial least squares, and artificial neural networks with sigmoidal and radial basis activation functions, that can be used to determine input-output relations. Connectivity methods are developed to facilitate data reduction and to determine significant input-output dependence. Comparison of the predictive abilities of these approximators and their performance is conducted using the data obtained from the DIII-D plasma fusion experiment.