xref: /petsc/src/ts/tutorials/power_grid/ex3sa.c (revision d71ae5a4db6382e7f06317b8d368875286fe9008)
1 
2 static char help[] = "Adjoint and tangent linear sensitivity analysis of the basic equation for generator stability analysis.\n";
3 
4 /*F
5 
6 \begin{eqnarray}
7                  \frac{d \theta}{dt} = \omega_b (\omega - \omega_s)
8                  \frac{2 H}{\omega_s}\frac{d \omega}{dt} & = & P_m - P_max \sin(\theta) -D(\omega - \omega_s)\\
9 \end{eqnarray}
10 
11 F*/
12 
13 /*
14   This code demonstrate the sensitivity analysis interface to a system of ordinary differential equations with discontinuities.
15   It computes the sensitivities of an integral cost function
16   \int c*max(0,\theta(t)-u_s)^beta dt
17   w.r.t. initial conditions and the parameter P_m.
18   Backward Euler method is used for time integration.
19   The discontinuities are detected with TSEvent.
20  */
21 
22 #include <petscts.h>
23 #include "ex3.h"
24 
25 int main(int argc, char **argv)
26 {
27   TS           ts, quadts; /* ODE integrator */
28   Vec          U;          /* solution will be stored here */
29   PetscMPIInt  size;
30   PetscInt     n = 2;
31   AppCtx       ctx;
32   PetscScalar *u;
33   PetscReal    du[2]    = {0.0, 0.0};
34   PetscBool    ensemble = PETSC_FALSE, flg1, flg2;
35   PetscReal    ftime;
36   PetscInt     steps;
37   PetscScalar *x_ptr, *y_ptr, *s_ptr;
38   Vec          lambda[1], q, mu[1];
39   PetscInt     direction[2];
40   PetscBool    terminate[2];
41   Mat          qgrad;
42   Mat          sp; /* Forward sensitivity matrix */
43   SAMethod     sa;
44 
45   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
46      Initialize program
47      - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
48   PetscFunctionBeginUser;
49   PetscCall(PetscInitialize(&argc, &argv, (char *)0, help));
50   PetscCallMPI(MPI_Comm_size(PETSC_COMM_WORLD, &size));
51   PetscCheck(size == 1, PETSC_COMM_WORLD, PETSC_ERR_WRONG_MPI_SIZE, "Only for sequential runs");
52 
53   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
54     Create necessary matrix and vectors
55     - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
56   PetscCall(MatCreate(PETSC_COMM_WORLD, &ctx.Jac));
57   PetscCall(MatSetSizes(ctx.Jac, n, n, PETSC_DETERMINE, PETSC_DETERMINE));
58   PetscCall(MatSetType(ctx.Jac, MATDENSE));
59   PetscCall(MatSetFromOptions(ctx.Jac));
60   PetscCall(MatSetUp(ctx.Jac));
61   PetscCall(MatCreateVecs(ctx.Jac, &U, NULL));
62   PetscCall(MatCreate(PETSC_COMM_WORLD, &ctx.Jacp));
63   PetscCall(MatSetSizes(ctx.Jacp, PETSC_DECIDE, PETSC_DECIDE, 2, 1));
64   PetscCall(MatSetFromOptions(ctx.Jacp));
65   PetscCall(MatSetUp(ctx.Jacp));
66   PetscCall(MatCreateDense(PETSC_COMM_WORLD, PETSC_DECIDE, PETSC_DECIDE, 1, 1, NULL, &ctx.DRDP));
67   PetscCall(MatSetUp(ctx.DRDP));
68   PetscCall(MatCreateDense(PETSC_COMM_WORLD, PETSC_DECIDE, PETSC_DECIDE, 2, 1, NULL, &ctx.DRDU));
69   PetscCall(MatSetUp(ctx.DRDU));
70 
71   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
72     Set runtime options
73     - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
74   PetscOptionsBegin(PETSC_COMM_WORLD, NULL, "Swing equation options", "");
75   {
76     ctx.beta    = 2;
77     ctx.c       = 10000.0;
78     ctx.u_s     = 1.0;
79     ctx.omega_s = 1.0;
80     ctx.omega_b = 120.0 * PETSC_PI;
81     ctx.H       = 5.0;
82     PetscCall(PetscOptionsScalar("-Inertia", "", "", ctx.H, &ctx.H, NULL));
83     ctx.D = 5.0;
84     PetscCall(PetscOptionsScalar("-D", "", "", ctx.D, &ctx.D, NULL));
85     ctx.E        = 1.1378;
86     ctx.V        = 1.0;
87     ctx.X        = 0.545;
88     ctx.Pmax     = ctx.E * ctx.V / ctx.X;
89     ctx.Pmax_ini = ctx.Pmax;
90     PetscCall(PetscOptionsScalar("-Pmax", "", "", ctx.Pmax, &ctx.Pmax, NULL));
91     ctx.Pm = 1.1;
92     PetscCall(PetscOptionsScalar("-Pm", "", "", ctx.Pm, &ctx.Pm, NULL));
93     ctx.tf  = 0.1;
94     ctx.tcl = 0.2;
95     PetscCall(PetscOptionsReal("-tf", "Time to start fault", "", ctx.tf, &ctx.tf, NULL));
96     PetscCall(PetscOptionsReal("-tcl", "Time to end fault", "", ctx.tcl, &ctx.tcl, NULL));
97     PetscCall(PetscOptionsBool("-ensemble", "Run ensemble of different initial conditions", "", ensemble, &ensemble, NULL));
98     if (ensemble) {
99       ctx.tf  = -1;
100       ctx.tcl = -1;
101     }
102 
103     PetscCall(VecGetArray(U, &u));
104     u[0] = PetscAsinScalar(ctx.Pm / ctx.Pmax);
105     u[1] = 1.0;
106     PetscCall(PetscOptionsRealArray("-u", "Initial solution", "", u, &n, &flg1));
107     n = 2;
108     PetscCall(PetscOptionsRealArray("-du", "Perturbation in initial solution", "", du, &n, &flg2));
109     u[0] += du[0];
110     u[1] += du[1];
111     PetscCall(VecRestoreArray(U, &u));
112     if (flg1 || flg2) {
113       ctx.tf  = -1;
114       ctx.tcl = -1;
115     }
116     sa = SA_ADJ;
117     PetscCall(PetscOptionsEnum("-sa_method", "Sensitivity analysis method (adj or tlm)", "", SAMethods, (PetscEnum)sa, (PetscEnum *)&sa, NULL));
118   }
119   PetscOptionsEnd();
120 
121   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
122      Create timestepping solver context
123      - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
124   PetscCall(TSCreate(PETSC_COMM_WORLD, &ts));
125   PetscCall(TSSetProblemType(ts, TS_NONLINEAR));
126   PetscCall(TSSetType(ts, TSBEULER));
127   PetscCall(TSSetRHSFunction(ts, NULL, (TSRHSFunction)RHSFunction, &ctx));
128   PetscCall(TSSetRHSJacobian(ts, ctx.Jac, ctx.Jac, (TSRHSJacobian)RHSJacobian, &ctx));
129 
130   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
131      Set initial conditions
132    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
133   PetscCall(TSSetSolution(ts, U));
134 
135   /*   Set RHS JacobianP */
136   PetscCall(TSSetRHSJacobianP(ts, ctx.Jacp, RHSJacobianP, &ctx));
137 
138   PetscCall(TSCreateQuadratureTS(ts, PETSC_FALSE, &quadts));
139   PetscCall(TSSetRHSFunction(quadts, NULL, (TSRHSFunction)CostIntegrand, &ctx));
140   PetscCall(TSSetRHSJacobian(quadts, ctx.DRDU, ctx.DRDU, (TSRHSJacobian)DRDUJacobianTranspose, &ctx));
141   PetscCall(TSSetRHSJacobianP(quadts, ctx.DRDP, DRDPJacobianTranspose, &ctx));
142   if (sa == SA_ADJ) {
143     /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
144       Save trajectory of solution so that TSAdjointSolve() may be used
145      - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
146     PetscCall(TSSetSaveTrajectory(ts));
147     PetscCall(MatCreateVecs(ctx.Jac, &lambda[0], NULL));
148     PetscCall(MatCreateVecs(ctx.Jacp, &mu[0], NULL));
149     PetscCall(TSSetCostGradients(ts, 1, lambda, mu));
150   }
151 
152   if (sa == SA_TLM) {
153     PetscScalar val[2];
154     PetscInt    row[] = {0, 1}, col[] = {0};
155 
156     PetscCall(MatCreateDense(PETSC_COMM_WORLD, PETSC_DECIDE, PETSC_DECIDE, 1, 1, NULL, &qgrad));
157     PetscCall(MatCreateDense(PETSC_COMM_WORLD, PETSC_DECIDE, PETSC_DECIDE, 2, 1, NULL, &sp));
158     PetscCall(TSForwardSetSensitivities(ts, 1, sp));
159     PetscCall(TSForwardSetSensitivities(quadts, 1, qgrad));
160     val[0] = 1. / PetscSqrtScalar(1. - (ctx.Pm / ctx.Pmax) * (ctx.Pm / ctx.Pmax)) / ctx.Pmax;
161     val[1] = 0.0;
162     PetscCall(MatSetValues(sp, 2, row, 1, col, val, INSERT_VALUES));
163     PetscCall(MatAssemblyBegin(sp, MAT_FINAL_ASSEMBLY));
164     PetscCall(MatAssemblyEnd(sp, MAT_FINAL_ASSEMBLY));
165   }
166 
167   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
168      Set solver options
169    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
170   PetscCall(TSSetMaxTime(ts, 1.0));
171   PetscCall(TSSetExactFinalTime(ts, TS_EXACTFINALTIME_MATCHSTEP));
172   PetscCall(TSSetTimeStep(ts, 0.03125));
173   PetscCall(TSSetFromOptions(ts));
174 
175   direction[0] = direction[1] = 1;
176   terminate[0] = terminate[1] = PETSC_FALSE;
177 
178   PetscCall(TSSetEventHandler(ts, 2, direction, terminate, EventFunction, PostEventFunction, (void *)&ctx));
179 
180   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
181      Solve nonlinear system
182      - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
183   if (ensemble) {
184     for (du[1] = -2.5; du[1] <= .01; du[1] += .1) {
185       PetscCall(VecGetArray(U, &u));
186       u[0] = PetscAsinScalar(ctx.Pm / ctx.Pmax);
187       u[1] = ctx.omega_s;
188       u[0] += du[0];
189       u[1] += du[1];
190       PetscCall(VecRestoreArray(U, &u));
191       PetscCall(TSSetTimeStep(ts, 0.03125));
192       PetscCall(TSSolve(ts, U));
193     }
194   } else {
195     PetscCall(TSSolve(ts, U));
196   }
197   PetscCall(TSGetSolveTime(ts, &ftime));
198   PetscCall(TSGetStepNumber(ts, &steps));
199 
200   if (sa == SA_ADJ) {
201     /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
202        Adjoint model starts here
203        - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
204     /*   Set initial conditions for the adjoint integration */
205     PetscCall(VecGetArray(lambda[0], &y_ptr));
206     y_ptr[0] = 0.0;
207     y_ptr[1] = 0.0;
208     PetscCall(VecRestoreArray(lambda[0], &y_ptr));
209 
210     PetscCall(VecGetArray(mu[0], &x_ptr));
211     x_ptr[0] = 0.0;
212     PetscCall(VecRestoreArray(mu[0], &x_ptr));
213 
214     PetscCall(TSAdjointSolve(ts));
215 
216     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n lambda: d[Psi(tf)]/d[phi0]  d[Psi(tf)]/d[omega0]\n"));
217     PetscCall(VecView(lambda[0], PETSC_VIEWER_STDOUT_WORLD));
218     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n mu: d[Psi(tf)]/d[pm]\n"));
219     PetscCall(VecView(mu[0], PETSC_VIEWER_STDOUT_WORLD));
220     PetscCall(TSGetCostIntegral(ts, &q));
221     PetscCall(VecGetArray(q, &x_ptr));
222     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n cost function=%g\n", (double)(x_ptr[0] - ctx.Pm)));
223     PetscCall(VecRestoreArray(q, &x_ptr));
224     PetscCall(ComputeSensiP(lambda[0], mu[0], &ctx));
225     PetscCall(VecGetArray(mu[0], &x_ptr));
226     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n gradient=%g\n", (double)x_ptr[0]));
227     PetscCall(VecRestoreArray(mu[0], &x_ptr));
228     PetscCall(VecDestroy(&lambda[0]));
229     PetscCall(VecDestroy(&mu[0]));
230   }
231   if (sa == SA_TLM) {
232     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n trajectory sensitivity: d[phi(tf)]/d[pm]  d[omega(tf)]/d[pm]\n"));
233     PetscCall(MatView(sp, PETSC_VIEWER_STDOUT_WORLD));
234     PetscCall(TSGetCostIntegral(ts, &q));
235     PetscCall(VecGetArray(q, &s_ptr));
236     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n cost function=%g\n", (double)(s_ptr[0] - ctx.Pm)));
237     PetscCall(VecRestoreArray(q, &s_ptr));
238     PetscCall(MatDenseGetArray(qgrad, &s_ptr));
239     PetscCall(PetscPrintf(PETSC_COMM_WORLD, "\n gradient=%g\n", (double)s_ptr[0]));
240     PetscCall(MatDenseRestoreArray(qgrad, &s_ptr));
241     PetscCall(MatDestroy(&qgrad));
242     PetscCall(MatDestroy(&sp));
243   }
244   /* - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
245      Free work space.  All PETSc objects should be destroyed when they are no longer needed.
246    - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - */
247   PetscCall(MatDestroy(&ctx.Jac));
248   PetscCall(MatDestroy(&ctx.Jacp));
249   PetscCall(MatDestroy(&ctx.DRDU));
250   PetscCall(MatDestroy(&ctx.DRDP));
251   PetscCall(VecDestroy(&U));
252   PetscCall(TSDestroy(&ts));
253   PetscCall(PetscFinalize());
254   return 0;
255 }
256 
257 /*TEST
258 
259    build:
260       requires: !complex !single
261 
262    test:
263       args: -sa_method adj -viewer_binary_skip_info -ts_type cn -pc_type lu
264 
265    test:
266       suffix: 2
267       args: -sa_method tlm -ts_type cn -pc_type lu
268 
269    test:
270       suffix: 3
271       args: -sa_method adj -ts_type rk -ts_rk_type 2a -ts_adapt_type dsp
272 
273 TEST*/
274