Page 231 - Neural Network Modeling and Identification of Dynamical Systems

222 6. NEURAL NETWORK SEMIEMPIRICAL MODELING OF AIRCRAFT MOTION

FIGURE 6.13 The coefficient C x (α,φ) for various values of φ according to (A) the data from [20]and (B)theerrorofits
for fixed q = 0 deg/sec and V = 150 m/sec (From [38], used with permission from Moscow Aviation
approximation E C x
Institute).
T
are controls. The values of the controls are con- tially observable, y(t) =[V (t),θ(t),q(t)] .The
∈[−25,25] deg, δ th ∈[0,1].The output of the system y(t) is affected by additive
strained; δ e act
training and test sets were generated accord- Gaussian noise with a standard deviation σ V =
ing to the procedure described in the previ- 0.01 m/sec, σ θ = 0.01 deg, σ q = 0.005 deg/sec.
ous section, with a sampling period of t = The semiempirical ANN model learning is
0.01 sec. The vector of state variables is par- performed by the algorithms described in Chap-