Page 116 - Distributed model predictive control for plant-wide systems
P. 116

90                            Distributed Model Predictive Control for Plant-Wide Systems


           change? In this section, the performance deviation on single-step horizon under the commu-
           nication failure is discussed. Because MPC takes a receding-horizon control policy in which
           the optimization is resolved online at each sampling time with updated measurements, it is
           reasonable to focus on the single-step horizon.
             To indicate the communication connection among agents, define a connection matrix
           E = (e ).
                ij
             All elements in the main diagonal of E are zeros and other elements in the nonmain diago-
           nal of E are 1. 0.1 denotes the communication connection between two agents existed and 0
           shows no communication connection. Under the ideal situation of a communication connec-
           tion without structural disturbance e = 1(i, j = 1, … , m, i ≠ j), the output prediction model and
                                        ij
           the Nash optimal solution of the ith agent at the time instant k can, respectively, be rewritten as
                                                m
                                               ∑
                  ̃ y  (k)= ̃ y  (k)+ A Δu  (k)+   e A Δu   (k), (i = 1, … , m)  (5.64)
                   i,PM     i,P0    ii  i,M         ij  ij  j,M
                                              j=1, j≠i
           and                    [                           ]
                                                  m
                                                 ∑        ∗
                        ∗
                     Δu   (k)= D      − ̃ y  (k) −   G Δu   (k) , (i = 1, … , m)  (5.65)
                        i,M      ii  i  i,P0          ij  j,M
                                                j=1, j≠i
           where G = EA = [G ] denotes the dot multiplication with
                           ij
                                    e    ···  e    A    A     ···  A
                              ⎡      12        1m⎤ ⎡  11  12        1m⎤
                               e 21      ···  e 2m  A 21  A 22  ···  A 2m
                              ⎢                 ⎥ ⎢                   ⎥
                          G =  ⎢                ⎥ ⎢                   ⎥
                              ⎢ ⋮    ⋮        ⋮ ⎥ ⎢ ⋮    ⋮     0    ⋮ ⎥
                              ⎢                 ⎥ ⎢                   ⎥
                              ⎣e    e    ···  0 ⎦ ⎣A    A     ···  A  ⎦
                               m1    m2             m1    m2        mm
                                       e A     ···  e A
                              ⎡         12  12       1m  1m⎤
                               e A 21          ···  e A 2m
                              ⎢                           ⎥
                                                     2m
                                21
                            =  ⎢                          ⎥
                              ⎢  ⋮        ⋮     0      ⋮  ⎥
                              ⎢                           ⎥
                              ⎣e A     e A     ···     0  ⎦
                               m1  m1   m2  m2
           Then the Nash optimal solution of the whole system under convergent computation is
                                     ∗
                                                 −1
                                  Δu (k)= (I-D ) [  (k)− ̃ y (k)]                (5.66)
                                     M         E          P0
           with
                                                −D e A       ···  −D e A
                                 ⎡                 11 12  12        11 1m  1m⎤
                                  −D e A   21                ···  −D e A  2m
                                 ⎢                                          ⎥
                                     22 21
                                                                    22 2m
                    D =−D G =    ⎢                                          ⎥
                            1
                      E
                                 ⎢     ⋮            ⋮                 ⋮     ⎥
                                 ⎢                                          ⎥
                                 ⎣−D   e A     −D   e A      ···            ⎦
                                    mm m1  m1     mm m2  m2
           In the following analysis, assume that the prediction horizon and the control horizon are equal,
           and the communication failure is confined within a stable region. To analyze system perfor-
           mance deviation, define a communication failure matrix T. The matrix T is a diagonal matrix
           or a block diagonal matrix. For a diagonal matrix, define the elements of its main diagonal as
   111   112   113   114   115   116   117   118   119   120   121