The recent developments in the neural control field have made wide use of nonlinear model based control techniques. However, these techniques are not robust with respect to model estimation errors, and global control performance is affected by the inaccuracy of the estimated model. In particular, the presence of annoying off-set errors is normally experienced. In a non-adaptive control setting, this problem has to be addressed by imposing structural constraints on the model or on the controller. On the other hand, conventional model based techniques are not suitable for structural modifications, since the controller structure is directly related to that of the estimated model by means of inversion or optimization techniques. Moreover, the design of a nonlinear controller is a demanding task, and ana posteriorimethod for the rejection of the error, which doesn't modify the already designed nonlinear controller, is highly desirable. In this paper, various error rejection methods and bias-free control structures are investigated, which introduce no modification to the independently designed nonlinear model based controller, and simply add on to the existing control structure.
展开▼
