PGscen

This package generates power grid scenarios using GEMINI trained on historical grid asset actual and forecasted values. Please see the Jupyter notebooks available in the examples/ directory for an overview of how PGscen works.

Installation

PGscen is available for installation on Linux-based operating systems such as Ubuntu as well as macOS. To install PGscen, first clone this repository at the latest release:

git clone https://github.com/PrincetonUniversity/PGscen.git

Next, navigate to the cloned directory to create and activate the conda environment containing the prerequisite packages for PGscen:

conda env create -f environment.yml

conda activate pgscen

From within the same directory, complete installation of PGscen by running:

pip install .

Running PGscen on Texas 7k

Once installed, you can generate scenarios using the NREL/ERCOT datasets for the Texas 7k system installed as part of the package. This is easiest done using the command line interface made available upon installation.

In particular, this package includes four command line tools: pgscen-load, pgscen-wind, pgscen-solar, and pgscen-load-solar, each of which create scenarios for the given asset type. While the first three consider load, wind, and solar assets separately, pgscen-load-solar models load and solar assets jointly to account for the fact that both are influenced by the same factors, and especially the weather. You can also use the pgscen command, which generates all three of load, wind, and solar scenarios at the same time.

The basic interface for all five of these tools is the same, with two required arguments:

• start: The starting date for scenario generation, given in YYYY-MM-DD format. Currently supported dates (all inclusive) are:
  • 2017-01-03 to 2018-12-31 for load scenarios
  • 2018-01-01 to 2018-12-30 for wind scenarios
  • 2017-01-04 to 2018-12-30 for solar scenarios, except for 2017-12-31 and 2018-01-01
  • 2017-01-05 to 2018-12-30 for load-solar scenarios, except for 2017-12-31 and 2018-01-01
• days: The number of days to create scenarios for. Note that the implied date range must also fall within the intervals listed above for start.

Unlike the other tools, pgscen takes the optional --joint argument, in which case it will generate load and solar scenarios jointly instead of the default behaviour of modeling them separately.

These tools also all take the following optional arguments:

• --out-dir (-o): Where to store generated scenarios. By default, output files are stored in the PGscen/data subdirectory of where this repository was cloned.
• --scenario-count (-n): How many scenarios to generate. By default, 1000 are created.
• --verbose: Whether to print messages on the state of the scenario generator. Use -v, -vv for escalating levels of verbosity.
• --pickle (-p): Instead of writing a separate .csv file for each asset's daily scenarios, save the generated scenarios for all assets for each day as a single Python pickle file
• --test: Run scenario generation using cached subsets of the ERCOT/NREL datasets to test if installation was carried out correctly.

Example usages include:

pgscen-load-solar 2017-06-29 2

pgscen-load-solar 2018-01-25 2 -p -v

pgscen-load 2018-03-02 2 -o scratch/output/PGscen

pgscen-wind 2018-07-05 3 -n 2000

pgscen 2018-05-11 3 -p

pgscen-solar 2017-09-22 1 -n 500 -vv

pgscen 2018-05-11 3 --joint -o scratch/output/PGscen -p

The implementations of these command line interfaces are located at pgscen/command_line.py, and can be used as templates for designing your own PGscen workflows. See for example pgscen/rts_gmlc/command_line.py for one such custom implementation which is described in greater detail below.

Running PGscen on RTS-GMLC

PGscen also includes a beta version of the pgscen-rts command line tool for generating scenarios for the RTS-GMLC power grid system. The data for this system is not included with the package but can be acquired by cloning the RTS-GMLC repository:

git clone https://github.com/GridMod/RTS-GMLC.git

The interface for pgscen-rts is identical to that of its Texas 7k counterparts, with the exception of having to specify the directory where the RTS-GMLC dataset is located. For example, if you have a directory called repositories within which you cloned the RTS-GMLC repo, you would run:

pgscen-rts 2020-04-03 4 repositories/RTS-GMLC

Note that this tool creates both joint load-solar and wind scenarios in one run; it also saves output as a single compressed Python pickle object for each day unless the --csv flag is given, in which case it reverts to the original format used by the Texas 7k tools. pgscen-rts supports any day from 2020-01-01 to 2020-12-31 inclusive.

Please see pgscen/rts_gmlc/create_scenarios.sh for an example of how this tool can be used to generate RTS-GMLC scenarios for the entire year using a Slurm compute cluster.