Direct Controller Design from Plant Data

Methods are investigated in which controllers (simple memoryless nonlinear functions of the output of the plant) are directly estimated from input-output data without a model of the plant. A procedure is proposed which first selects good data sequences according to some control objective and then estimates the controller parameters directly from the selected sequences. Connections are drawn with other methods of direct controller design, e.g. fuzzy logic control or artificial neural networks control and general issues regarding these methods, such as a priori knowledge and experiment design, are considered.