The use of chaos observation as a performance measure in neurocontrol using forward plant dynamics

In neurocontrol using forward plant dynamics it is important to detect improperly trained areas. If the plant is not generalized at an operating point then the control action shows unpredictable behavior. A method based on Poincare´s map is introduced to detect the improperly trained areas and tested using simulations. It is observed that the chaotic and oscillatory actions can be detected. The neurocontroller leaves the plant to a more traditional controller if chaos is detected, and changes the neurocontroller parameters if oscillatory behavior is detected.