Neural Hybrid Control

Abstract

There have been many applications of neural networks (NN) in controls. Algorithms that effectively tune the weights on-line have been developed. NN applications in control can be broadly classified into two sorts: identification and control. In this chapter, it is intended to use NN for closed-loop control of a system with a dead-zone sandwiched in between two dynamic blocks. An adaptive version of the hybrid control scheme presented in Chapter 4, will be developed using a NN to compensate the unknown sandwiched dead-zone. As the controller structure is hybrid and has neural networks, we term it as a neural-hybrid controller. The proposed neural-hybrid controller consists of an inner loop discrete-time feedback structure incorporated with an adaptive inverse using NN for the unknown nonlinearity N(·), in the present case, a dead-zone, and an outer-loop continuous-time feedback control law for achieving desired output tracking [86]. The dead-zone compensator consists of two NN’s, one used as an estimator of the sandwiched dead-zone function and the other for the compensation itself. To approximate jump functions such as a dead-zone inverse, it is found that the NN that uses smooth activation functions should be augmented with extra nodes containing a jump function approximation basis set of discontinuous activation functions. This is necessary as otherwise to approximate such jump functions using smooth activation functions, many NN nodes and many training iterations are required [40], [60].