LT Plan At a Time
| Units: | - |
| Default Value: | 0 |
| Validation Rule: | ≥-1 |
| Description: | Number of years solved in each step of LT Plan |
| Detail: |
The default behavior of LT Plan is to optimize the entire planning horizon in one 'step', and this is the ideal solution. However, for very large models it might be necessary to break the optimization into multiple steps or to resolve the horizon progressively using Rolling Horizon. The method used is controlled by the Overlap setting combined with At a Time.
- Multi-step
-
For the multi-step method, this option controls the number of years solved in each step. A value of '0' means that the entire horizon is solved in one step.
Note that you must select a number of years that divides the years in the planning horizon precisely. For example, the following combinations are valid:
- Solve 30 years in 3 steps of 10 years
- Solve 21 years in 3 steps of 7 years
The following are not valid:
- Solve 30 years in 4 steps
- Solve 20 years in 3 steps
- Rolling Horizon
- For the Rolling Horizon method, this option decides the length of
each 'roll'. For example if you set At a Time = 5 then each solve
will have full temporal resolution for five years and rolling will
occur every four years (one year is discarded from each roll by
default). For a 20 year horizon with At a Time = 5 the following
rolling steps are performed:
- Years 1-20 with full detail in years 1-5 and less detail in years 6-20
- Years 5-20 with full detail in years 5-9 and less detail in years 10-20 and optimal build decisions for years 1-4 from the previous solution
- Years 10-20 with full detail in years 10-14 and less details in year 15-20 and optimal build decisions for years 1-9 from the previous solutions
- Years 15-20 with full detail in years 15-19 and less detail in year 20 and optimal build decisions for years 1-14 from the previous solutions
- Year 20 with full detail and optimal build decisions for years 1-19 from the previous solutions
- A final full detail run with all build decisions locked.
Note that these methods work independently of the Chronology setting, meaning you can further reduce the computational burden of the simulation by choosing Partial, Fitted or Sampled chronologies.