Page 178 - Neural Network Modeling and Identification of Dynamical Systems
P. 178

5.1 SEMIEMPIRICAL ANN-BASED APPROACH TO MODELING OF DYNAMICAL SYSTEMS  169
                          where ˆ x:[0, ˜ t]→ R n x  is a solution of the initial value  that
                          problem for the system of ODEs with right hand
                              ˆu
                                         ˆ
                                                                                                   ∗
                                                                                                          ∗
                                                                                         ∗
                                                                                   ˆ
                                                                               ∗
                                                                                     ∗
                                                                            ∗
                          side f (t, ˆ x(t)) ≡ f(ˆ x(t),u(t)) and initial condition   f(x ,u ) − f(x ,u )  <ε f  ∀x ∈ X ∀u ∈ U.
                                 s
                          ˆ x(0) = ˜ x ; y(t) = g(x(t)); ˆy(t) = ˆ g(ˆ x(t)).
                                                                         Next, we estimate the upper bound on the ap-
                          Proof. For any vector-valued functions f ∈ F and
                                                                       proximation error for the right hand side of the
                          f ∈ F, the vector-valued functions f u  and f ˆu
                              ˆ
                          ˆ
                                                                       system of ODEs, evaluated at each point of the
                          satisfy all conditions of the theorem on exis-  true state space trajectory x(t) on the whole time
                          tence and uniqueness of solutions to initial value
                                                                       segment. We have
                          problems (see Theorem 54 in [16]), due to the fol-
                          lowing considerations:
                                                                               t

                                                                                 u
                                                                                           ˆ u
                          • the vector-valued function u(t) is measur-   sup     f (s,x(s)) − f (s,x(s))ds

                            able, while vector-valued functions f(x,u)  t∈[0,˜ t]
                                                                              0
                            and f(ˆ x,u) are continuous with respect to           t
                                 ˆ
                                                         u
                                                              ∗
                            u; therefore the compositions f (t,x ) and
                                                                                               ˆ u
                                                                                     u

                                                                            sup     f (s,x(s)) − f (s,x(s))  ds
                            ˆu   ∗
                            f (t, ˆ x ) are measurable with respect to t for
                                                                            t∈[0,˜ t]
                            all ˆ x ;                                            0
                                ∗
                          • the vector-valued function u(t) is locally inte-     ˜ t                        ˜ t
                                                                                 u         ˆ u              f
                            grable, while vector-valued functions f(x,u)  =     f (s,x(s)) − f (s,x(s))  ds <  ε ds
                                ˆ
                            and f(ˆ x,u) are locally Lipschitz continuous    0                           0
                            with respect to u; therefore the compositions     f
                                                                          = ˜ tε .
                             u
                                        ˆu
                                              ∗
                            f (t,x ) and f (t, ˆ x ) are locally integrable
                                 ∗
                            with respect to t for all ˆ x ;
                                                 ∗
                                                                         According to the assumption of this theo-
                                                           ˆ
                          • vector-valued functions f(x,u) and f(ˆ x,u) are  rem, the initial condition ˜ x is contained in the
                                                                                               s
                            continuous and locally Lipschitz continuous  δ-neighborhood of x , i.e.,
                                                                                         s
                            with respect to x; therefore the vector-valued
                                                 ˆu
                                      u
                                        ∗
                            functions f (t ,x) and f (t , ˆ x) are also con-            s   s
                                                    ∗
                                                                                       x − ˜ x   <δ.
                            tinuous and locally Lipschitz continuous with
                            respect to x for all t .
                                             ∗
                                                                       Also, we have assumed that all vector-valued
                                                                                             ˆ
                                                                                ˆ
                                                                                    ˆ
                          Thus, there exists the unique solution x of the  functions f ∈ F and ˆ g ∈ G satisfy the Lipschitz
                          initial value problem for the system of ODEs  condition, i.e.,
                          with right hand side f u  and initial condition

                                 s
                          x(0) = x ,aswellasthe unique solution ˆ x of    ˆ  	  ∗  ˆ  		  ∗    f   	  		    ∗



                                                                                                x − x , ∀u ∈ U,
                                                                        f(x ,u ) − f(x ,u )   M
                                                              ˆu
                          the similar problem with right hand side f and
                                                                                	            g

                                                s


                          initial condition ˆ x(0) = ˜ x . According to the as-    ˆ g(x ) − ˆ g(x )   M  x − x , ∀x ,x ∈ X,
                          sumption of this theorem, the solution x exists
                          on the whole segment [0, ˜ t] and is contained in  for some nonnegative real Lipschitz constants
                                                                         f
                                                                                g
                          X along with the closure of its ε-neighborhood.  M and M . Let us also assume that
                            Since the set of vector-valued functions F is
                                                                 ˆ
                                                                                          ε      f
                          everywhere dense in the space F,for anyvector-             δ =    g  e −M ¯ t  ,
                          valued function f ∈ F and any positive real ε f                3M
                                                                                                 f
                                                                                     f     ε   −M ¯ t
                          there exists a vector-valued function f ∈ F such          ε =     g  e   .
                                                           ˆ
                                                              ˆ
                                                                                         3M ¯ t
   173   174   175   176   177   178   179   180   181   182   183