Page 124 - Neural Network Modeling and Identification of Dynamical Systems
P. 124

3.4 ANN-BASED CONTROL OF DYNAMICAL SYSTEMS               113

































                          FIGURE 3.10 Comparison of the network operation with the preprocessing layer (with the stabilizer actuator model) and
                          the mathematical model (3.23). The solid line is model (3.23) output; the dotted line is the output of the neural network
                          model; V z is the component of the velocity vector along the Oy-axis; q is the angular velocity of the pitch; α is the angle of
                                                                                                    (ref )   (ref )
                          attack; δ e is the deflection angle of the stabilizer; t is the time; EV z , Eq,and Eα are the differences |V z − V z  |, |q − q  |,
                          and |α − α (ref ) |, respectively (From [99], used with permission from Moscow Aviation Institute).

                          3.4.2.1 Basic Approaches to the Use of       third approach (“compromise”) is a combination
                                 ArtificialNeuralNetworksin             of a conservative and radical approach, or more
                                 Control Problems                      precisely, some compromise between them.
                                                                         In the general case, the most effective (in
                            Thereare threemainapproaches totheuse of
                                                                       terms of the applied problem) will, of course,
                          ANNs in control systems [83–86].
                                                                       be a radical approach. However, a less pow-
                            In the first approach (“conservative”) the  erful conservative approach not only has the
                          structural organization of the control system re-
                                                                       right to exist, but, moreover, is more prefer-
                          mains unchanged, i.e., as it was obtained when
                                                                       able at present for a variety of reasons, both
                          designing such a system using one of the tradi-
                                                                       objective and subjective. Namely, following the
                          tional approaches. In this case, the ANN plays  conservative approach, it is possible to quickly
                          the role of a module for correcting specific pa-  achieve practically meaningful results, since this
                          rameters of the control system (for example, its  approach is not about creating a new control sys-
                          gains) depending on the operating conditions of  tem from scratch, but about upgrading existing
                          the system. In the second approach (“radical”),  systems. Further, it is very difficult to imagine
                          the whole control system or a functionally com-  now the situation when on board (for example,
                          pleted part is realized as an ANN system. The  an airplane) a control system based entirely on
   119   120   121   122   123   124   125   126   127   128   129