Page 223 - Neural Network Modeling and Identification of Dynamical Systems
P. 223
214 6. NEURAL NETWORK SEMIEMPIRICAL MODELING OF AIRCRAFT MOTION
FIGURE 6.9 Evaluation of the generalization ability of the ANN model after the final 1000-step learning phase: E α , E β , E p ,
E r , E q are the prediction errors for the corresponding observable variables; the straight lines on the three upper subgraphs
show the values of the control variables corresponding to the test maneuver (From [4], used with permission from Moscow
Aviation Institute).
as evenly and densely as possible the entire state der to authentically reflect in the ANN model
space of the system), as well as the variety of the dynamics of the simulated system). An addi-
changes in the neighboring states in time (in or- tional complicating factor is that the subsequent