Incremental Model Reference Adaptive

In this paper we are describing and illustrating the Incremental Model Reference Adaptive Polynomial Controllers Network (IMRAPCN). This algorithm is a polynomial version of the conventional linear controllers network. Two important properties of that system are: (1) its behaviour is clearly understandable because each polynomial controller can be interpreted in linear terms and (2) it is capable of an autonomous construction of its structure which involves the clustering of the validity space. We illustrate its control capability according to the control of two highly non linear systems. The results show that the IMRAPCN enables a very accurate control of the non linear systems over a wide operating range. This system can be used for the control of a possibly discontinuous non linear system and it is not aaected by the \stability-plasticity dilemma". The use of polynomial controllers network enables a much more accurate control of a non linear system than the one that can be obtained by using linear controllers network. Finally, a polynomial controllers network of low order (e.g. cubic polynomial) is much more exible than a single polynomial controller. This exibility could make the polynomial controllers network adaptable to any kind of system. From these features the IMRAPCN appears very promising as a powerful and general algorithm for the control of non linear systems.