ST Schedule Class
| Description: | ST Schedule simulation phase |
| Detail: |
See also ST Schedule Property Reference for a detailed list of properties for this class of object.
Contents
1. Introduction
ST Schedule is mixed-integer programming (MIP) based chronological optimization. It is distinct from LT Plan and MT Schedule in that it model days of the horizon at full resolution, as dictated by the Horizon Periods per Day setting. At the default setting this means every hour, but the resolution can be customized to any feasible length e.g. 5-minute intervals.
ST Schedule is designed to emulate the dispatch and pricing of real market-clearing engines, but it provides a wealth of additional functionality to deal with:
- unit commitment
- constraint modelling
- financial/portfolio optimization
- Monte Carlo simulation
- stochastic optimization
Emulation of real market-clearing engines involves clearing generator offers against forecast load accounting for transmission and other constraints to produced a dispatch and pricing outcome. ST Schedule can do this but the simulator extends this basic functionality by allowing you to specify fundamental data such as generator start costs and constraints, heat-rate curves, fuel costs, etc. as well as or in addition to market data such as generator offers, and the dynamic formulation engine in the AMMO software at the heart of the simulator tailors the representation of each simulation element, such as a generator, at runtime and on a case-by-case basis. This allows you to seamlessly mix market data with fundamental data as desired - relying on the simulator to compute the appropriate market representation at runtime, and maximize simulation efficiency.
2. ST Schedule Chronology
ST Schedule provides two methods for modelling the chronology:
- Full Chronology
- Every dispatch interval inside the ST Schedule horizon is modelled.
- Typical Week
- One selected week is modelled each per month in the horizon and results are applied to the other weeks.
2.1. Full Chronology
In this mode, ST Schedule runs every trading period and maintains chronological consistency across the horizon. For example it can model generator start ups and shutdowns and track the status of units across time. The Horizon options allow you to select either the whole or only a subset of the planning horizon for execution with ST Schedule.
When selecting the planning horizon, the step type is chosen from years, months, weeks, or days. But the ST Schedule Step Type must be either weeks, days, hours, or minutes. The reason for this is related to the way in which PLEXOS sets up and solves the ST Schedule problem. At runtime PLEXOS:
- dynamically constructs a mathematical programming problem to represent the first step of the ST Schedule; and then
- as each step is evaluated, the same problem is simply modified to represent the next step, and so on until the required horizon has been evaluated.
The length of each ST step is controlled by the ST Schedule At a Time property. In general the longer each step of the ST Schedule is, the greater the execution time will be for each of those steps, but there is some overhead is switching from one step to the next. You should experiment with these settings to find the best combination for their models.
Note further, that the outcome of the simulation can be influenced by the size of the ST step when there are significant intertemporal aspects. This is because the state of the system, e.g. generator unit commitment, is recorded and carried over from one step to the next, but each step does not look-ahead to the next. Hence unless the model has no intertemporal elements, e.g. when performing a pure market-clearing emulation, it is recommended that the ST be run in steps of no less than one day at a time.
To further improve the optimization of unit commitment decisions you can configure ST Schedule to use a look-ahead period ahead of each step. This allows the step size of ST Schedule to be kept small ( e.g. a day at a time) but sufficient look-ahead maintained for unit commitment decisions.
2.2. Typical Week
When ST Schedule is run in typical week mode, the horizon options are simplified. The simulation will always run across the whole planning horizon, and the only option to chose is the size of each step of the ST Schedule i.e. how many trading periods should be 'solved' at once. Typically, large models should be run in daily or evenly hourly steps, smaller models can run in weekly steps.
Running in this mode reduces the amount of simulation work for ST Schedule by more than a factor of four, but PASA and MT Schedule are still run in exactly the same manner.
Note that solution data for the typical week is written into the solution database, and the user interface will explode those data out so that a full chronology can be viewed. This means that the size of the solution database is also reduced when ST Schedule is run in this mode. Summary data (daily, weekly, monthly, annual) are all calculated based on the mapping of typical weeks to trading periods i.e. the daily data for a day that was not part of a typical week, is taken from the day of the same type in the typical week that was run. Thus, summary data are complete, but may not match input data such as total energy, and peak demand.
The selection of week inside the month is controlled by the option Typical Week. The beginning day of week is set by the Week Beginning option. When set to automatic, the week begins on the same day of week as the first day of the planning horizon.
3. Step Link Mode
By default the steps of ST Schedule are executed in sequence and each links to the previous for initial conditions. Step link behaviour is controlled with the setting Step Link Mode and additional options. Steps can be executed in parallel for a significant speed improvement and to better utilized compute resources.
4. Multi-sample Simulation
4.1. Monte Carlo Simulation
ST Schedule is capable of performing Monte Carlo simulation where the samples differ in:
- the outage state of Generators and/or Lines
- the value of one or more stochastic Variables
The number of samples executed is controlled by settings on the Stochastic object associated with the executing Model.
The Stochastic Method setting allows you to choose between running the samples sequentially, or in parallel. Parallel is more efficient but requires more memory to execute.
Guided Monte Carlo is a variation in which only selected simulation steps solve all samples. This is available via a hidden parameter.
4.2. Stochastic Optimization
The Stochastic Method setting also allows the stochastic samples to be used as 'scenarios' in a two-stage stochastic optimization, where selected decisions are taken as 'non-anticipative' while others are recourse. For example, this approach can determine the optimal unit commitment of selected generators given uncertainty in forecast load or price. See the Unit Commitment Non-anticipativity for more information.
4.3. Reporting
Reporting can be done for every sample and additionally statistics can be reported (min, max, mean, std. dev.). See the Report topic for more details.