Recent advances on dynamic learning from adaptive NN control
Min WANG, Cong WANG
Notice that intelligent control is developed based on learning and control abilities of human
beings. As such, an ideal intelligent control system should not only have the ability to adapt
to complex dynamic environments, but also the ability to imitate human learning. The learning
ability means to acquire the knowledge of uncertain closed-loop dynamics from dynamic
environments, and use the stored experience knowledge to improve control performance. Such an
autonomous learning has been viewed as one of the most important and indispensable capabilities
to achieve a high degree of system autonomy. The existing intelligent control themes, including
adaptive neural control and adaptive fuzzy control, are usually exclusively on the system
stability by the use of online adjustment parameters to estimate the unknown dynamics, in which
the exponential convergence of the estimated parameters is not a major concern. As a result,
most of existing intelligent control schemes still need to readjust the controller parameters
for the same or similar control tasks. In this sense, as shown in, the learning ability of
intelligent control is actually very limited and still needs further research. However,
it is very challenging to acquire and express the knowledge of unknown dynamics in a
nonstationary or dynamic environment.