Stochastic Deep Branching

Units:-
Default Value:0
Validation Rule:≥0
Description:For Rolling Horizon this is the level of additional branching after the end of regular branching

Stochastic Deep Branching applies to the Rolling Horizon method for solving multi-stage stochastic optimization - see MT Schedule Stochastic Algorithm.

During iterations of the Rolling Horizon, branching in the scenario tree occurs up to a given number of stages (weeks/months) ahead. That number of stages is controlled by the Minimum Sample Weight setting. In the following example, branching occurs for six stages ahead each roll (iteration) of the solution method:

        Started Multi-stage Stochastic Optimization
        --------------------------------------------------------------------------------------------------------------------------------
        Tasks: 87   Cols 49,868,052   Ints 0   Rows 67,006,443   Nzs  140,512,308
        --------------------------------------------------------------------------------------------------------------------------------
        Coefficient statistics:
          Matrix Range:     [-1.006585e+008] - [9.192960e+004]    [9.447895e-004] - [1.006585e+008]
          Objective Range:  [-8.095759e+001] - [1.376994e+008]    [4.767986e-002] - [1.376994e+008]
          Bounds Range:     [5.952380e-003] - [2.394257e+006]    [5.952380e-003] - [2.394257e+006]
          RHS Range:        [-4.270813e+007] - [1.509025e+003]    [3.157895e-002] - [4.270813e+007]
        --------------------------------------------------------------------------------------------------------------------------------
        Started Roll 1 of 38. Branching Stages 5-9 of 157
        Completed Roll 1 of 38. Time: 00:06:24.3. Elapsed: 00:11:55.0. Remaining: 03:40:28.4.
        Started Roll 2 of 38. Branching Stages 9-13 of 157
        Completed Roll 2 of 38. Time: 00:06:18.4. Elapsed: 00:18:13.5. Remaining: 03:38:42.4.
        Started Roll 3 of 38. Branching Stages 13-17 of 157
        Completed Roll 3 of 38. Time: 00:05:58.8. Elapsed: 00:24:12.3. Remaining: 03:31:48.2.
        Started Roll 4 of 38. Branching Stages 17-21 of 157
        Completed Roll 4 of 38. Time: 00:06:04.7. Elapsed: 00:30:17.1. Remaining: 03:25:56.2.
        ...
        Completed Roll 37 of 38. Time: 00:04:18.5. Elapsed: 03:16:49.2. Remaining: 00:05:10.7.
        Started Roll 38 of 38. Branching Stages 153-157 of 157
        Completed Roll 38 of 38. Time: 00:04:06.7. Elapsed: 03:20:55.9. Remaining: 00:00:00.0.
        Started Deterministic Phase
        Solving 87 Branches in Parallel Batches of 87
        Solving Branches 1-87 of 87
        Completed Branches 1-87 of 87. Time: 00:20:08.6. Elapsed: 00:20:08.6. Remaining: 00:00:00.0.
        Completed Deterministic Phase. Time: 00:20:08.6
        Completed Multi-stage Stochastic Optimization. Time: 03:41:51.4
        MT Schedule Step 1 of 1. Time: 03:44:09.4. Elapsed: 03:44:10.5. Remaining: 00:00:00.0.

After six stages branching ceases for that iteration i.e. each full branch is deterministic from that stage onward. When the next roll (iteration) begins a further six stages of branching are added and so on as the algorithm rolls through the horizon.

Deep branching allows for additional branching after the regular branching ceases. The default value of zero means no additional branching. Values above zero space additional branching at factors of 2 from the end of regular branching. For example, Deep Branching = 4 results in additional branching 2, 4, 8 and 16 stages after regular branching. Deep Branching improves the ability of Rolling Horizon to 'look-ahead' and may improve solutions in cases where the system is highly sensitive to the stochastic variables e.g. systems with a high dependency on hydro.