Page 171 - Neural Network Modeling and Identification of Dynamical Systems
P. 171

4.3 APPLICATION OF ANN MODELS TO ADAPTIVE CONTROL PROBLEMS UNDER UNCERTAINTY CONDITIONS  161
                          on the system which deviates the trajectory of  was obtained initially, without performing the
                          the real object from the reference trajectory. We  adaptive adjustment, only due to the robustness
                          try to reduce this deviation embedding a com-  properties (see Figs. A.79–A.81 for the MRAC
                          pensating loop (compensator) into the system.  system and Figs. A.82–A.84 for the MPC sys-
                          As showninSection 4.3.2.3, the embedding of  tem).
                          the compensating loop into the adaptive sys-   In the conducted experiments, the results of
                          tem increases its robustness, without which the  which are presented in Figs. A.79–A.84, the sys-
                          adaptation algorithm may not meet the adjust-  tem operation time was divided, as in the previ-
                          ment of the control law in that time window  ous case, into two segments of 20 sec each. The
                          within which the situation must not have time  nature of the control action concerning the an-
                          to grow from an emergency to a catastrophe.  gle of attack is the same as in the last case, i.e.,
                          In a number of cases, changes in the dynam-  the first 20 sec command signal is a random se-
                          ics of the controlled object and/or the condi-  quence of frequently changing values of the an-
                          tions of its operation are not significant. In these  gle of attack. With such a signal, the disturbance
                          cases, its robustness will be sufficient for the nor-  from the input signal does not have enough time
                          mal operation of the system, and it is possible  to decay, as the next signal acts on the object.
                          not to use the adaptation mechanism. For this  Such a complicated signal makes it possible to
                          reason, questions arise about the importance of  estimate the cumulative effect of several distur-
                          the mechanism of adaptation. We need to know  bances whose responses intersect in time, i.e.,
                          what this mechanism provides in comparison   it allows us to test the control system in fairly
                          with the version of the system without adapta-  harsh conditions. Unlike the previous case, the
                          tion, but with robustness improved by using a  adjustment of the control law was not carried
                          compensating loop.                           out, that is, the adaptation mechanism was dis-
                            The experiments were carried out to assess  abled.
                          the contribution of adaptability and robust-   In the second time interval, the control system
                          ness to the behavior of the system. We deac-  is tested using a more traditional input signal,
                          tivate the adaptation mechanisms in these ex-  which makes it possible to evaluate the behav-
                          periments. As in the previous case (analysis  ior of the system with isolated stepwise distur-
                          of the influence of uncertainties in the source  bances.
                          data on the properties of the system being syn-  The results demonstrate that the inaccuracy
                          thesized), the system was initially set up in-  of the adjusting is not critical for the stability
                          correctly. In the previous series of experiments  of the system, but causes significant deviations
                          described in Section 4.3.4.2, the control sys-  from the reference trajectory. In a steady state,
                          tem was adapted to the changed conditions for  these deviations can be reduced to zero using
                          the operation of the controlled object. For this  an integral compensator, but in transient modes,
                          purpose, in the experiments whose results are  this cannot be handled, and the only way to im-
                          showninFigs. A.61–A.69 (MRAC scheme) and     prove accuracy is to activate adaptation mecha-
                          Figs. A.70–A.78 (MPC scheme), we set up for  nisms to reduce the error below some prescribed
                          20 sec the disturbing reference signal required  level.
                          to adjust the control law, and then the test step-  As we can see from the results presented in
                          wise signal for estimating the tuning quality. In  Figs. A.79–A.84, the lack of mechanisms of adap-
                          a series of experiments, the results of which are  tation leads to a sharp deterioration in the qual-
                          presented below, the control system continued  ity of control. Namely, the error of tracking the
                          to operate with this “incorrect” setting, which  reference signal with respect to the angle of at-
   166   167   168   169   170   171   172   173   174   175   176