Input data of an environment¶
A grid2op “environment” is nothing more than a local folder on your computer.
This folder consists of different things:
a file representing a powergrid (for available environment at time of writing grid.json) and a “model” to make sense of this file (usable in grid2op thanks to a Backend)
FOCUS OF THIS FILE some “input time series” for the generation and the load
some other optional files, such as:
the characteristics of loads
of storage units
or the “geographical” coordinates of the elements of the grid for example.
Role of these time series¶
Grid2op model the real time behaviour of a powergrid. It supposes that somehow the amount of power consumed by each consumers is known as well as the amount of power each generator will produce.
In the case of grid2op, we suppose that all of these are “known” by the environment but unknown to
the agent. More specifically, the “agent” / “actor” / “controler” knows
(thanks to the
grid2op.Observation.BaseObservation) the amount of power produce at
a given time but not exactly what will be produced / consumed in the future. This is one of
the reason why “l2rpn environments” are “partially available” environments.
Currently, the package used to generate these time series is called chornix2grid_github but anything that generate data for all loads / generators could be used as “input” for a grid2op environment.
There are differences between the “theorical time series” generated that serve as input to a grid2op environment and the one the agent will see. Among the differences:
redispatching (a type of action performed by an agent) will modify the production of controlable generation units such as thermal, nuclear or hydro
curtailment (another type of action performed by an agent) can decrease the production of generator using renewable energy sources
storage unit (another type of action) will have some impact on the generators. A fondamental “law” of energy grid is that it should be balanced at all time. If a storage units inject on the grid xxx MW then “something else” should decrease its production of xxx MW. In grid2op we suppose that this “something else” is controlable generators.
modification of topology or any other type of actions (and sometimes even “non action”) will impact the electrical losses on the grid. And because the “fondamental law of energy grid” (power injected should match exactly power removed and power lost) a set of special generators (they have a fancy name: “generators participating to the slack bus”) are modified by the
Depending on the environment, some “forecast” are available. Forecasts represent a view at a given time for the near future. In most environment the forecast are available for the next step.
You can however, thanks to Time Series Handlers “generate” forecasts for longer horizons if you need them.
Different type of “data”¶
To run a grid2op environment, you need to provide different type of data to the
the data includes:
“load_p”: active consumption for all loads at all steps of the environment (in MW)
“load_q”: reactive consumption for all loads and at all steps of the environment (in MVAr)
“gen_p” / “prod_p”: active production for all generators at all steps of the environment (in MW)
“gen_v” / “prod_v”: voltage setpoint for each geneators at all steps of the environment (in kV)
(optional) “load_p_forecasted”: some forecast of “a few” steps ahead for “load_p” (in MW)
(optional) “load_q_forecasted”: some forecast of “a few” steps ahead for “load_q” (in MVAr)
(optional) “prod_p_forecasted”: some forecast of “a few” steps ahead for “prod_p” (in MW)
(optional) “prod_v_forecasted”: some forecast of “a few” steps ahead for “prod_v” (in kV)
(optional) “maintenance”: whether some powerlines will be out of service for maintenance operations
***_forecasted will be the data used by the
or the or the
On some environment, if you dont’ want to use
grid2op.Observation.BaseObservation.get_forecast_env() and want to speed up
the time of the “step” function, you can call
The main classes you can use are:
If you want to read this data but also to read some forecasts (“load_p_forecasted”, “load_q_forecasted”, “prod_p_forecasted”, “prod_v_forecasted”) for next steps, you can use
If you want also to generate maintenance data “on the fly” (that varies between episode then) you might want to see
But if you want to deactivate the maintenance even if some files describing the maintenance are present in your “chronics” folder, have a look at:
Previous classes will rely on data being generated and available on the hard drive. This means than the same data will be used over and over again, you might want to consider Generating new data on the fly
grid2op.Chronics.FromChronix2grid[NB we DO NOT recommend to do that, it might takes a few minutes at each call to
grid2op.Environment.Environment.reset()and dramatically slow down your training]
And if you want some more control on what is being done, you can have a look at the dedicated module: Time Series Handlers
Example of such times series “generator”¶
If you still can’t find what you’re looking for, try in one of the following pages:
Still trouble finding the information ? Do not hesitate to send a github issue about the documentation at this link: Documentation issue template