Live forecast for the OCF nowcasting project
Project description
Nowcasting Forecast
Making live forecasts for the nowcasting project. The aim is to produce Grid Supply Point (GSP) Solar generation Forecasts. This is done by loading live PV results, Numerical Weather Predictions (NWPs) and Satellite images, and running this data through various different ML models.
You can run the application locally by:
python nowcasting_forecast/app.py --db-url='sqlite:///test.db'
Installation with conda
mamba env create -f environment.yml
conda activate nowcasting_forecast
# Installing fastai inside environment.yml is currently broken. So install separately:
mamba install -c fastchan fastai
# You need to manually install nowcasting_dataset, nowcasting_datamodel,
# nowcasting_dataloader, and power_perceiver. This can be done via pypi (for all except
# power_perceiver) or by git cloning the repo and using:
pip install -e <path_to_directory>
# Install nowcasting_forecast
pip install -e .
# You may also want to install dev tools:
mamba install pytest flake8 black pre-commit pydocstyle isort mypy
pre-commit install
Directories and files
The following shows the main files
+-- nowcasting_forecast
| +-- config
| +-- mvp_v0.yaml
| +-- mvp_v1.yaml
| +-- models
| +-- nwp_solar_simple.py
| +-- nwp_solar_simple_trained
| +-- model.py
| +-- nwp_simple_trained.py
| +-- app.py
| +-- batch.py
| +-- dataloader.py
+-- scripts
+-- tests
☀️ nowcasting_forecast
app.py
is the main entry point for running the forecast module. It contains the following arguments:
- --db-url: the database url that the forecasts will be saved too
- --fake: Create fake forecast, mainly used for testing
The app has three main steps
- Make batched data
- Run forecasting model to make forecasts
- Save forecasts to the database
batch.py
is used to load the data and create batches of data. This uses mainly used ManagerLive
dataloader.py
is used to load the batched data in an efficient way. This is current a stripped down version of this dataloader.
The config
directory contains configurations that are used to load different data sources.
The configuration is used by batch.py
to convert large amounts of data into batched data, which is then ready for ML models.
The database
directory contains database models and functions to interact with the database. See below for more information
The model
directory contains ml models to make forecasts. See below for more information
Models
NWP Simple
This takes the average 'dswrf' for each example and then divides this by 10 to give a rough model for MW per GSP
configuration: mvp_v0.yaml
NWP Simple trained
CNN model of 'dswrf' channel in NWP. 6 CNN layers then 4 fully connected layers.
Training run: https://app.neptune.ai/o/OpenClimateFix/org/predict-pv-yield/e/PRED-951/charts
configuration: mvp_v1.yaml
CNN
TODO
🩺 Testing
Tests are run by using the following command
docker-compose -f test-docker-compose.yml run forecast
These sets up postgres
in a docker container and runs the tests in another docker container.
This slightly more complicated testing framework is needed (compared to running pytest
)
as some queries can not be fully tested on a sqlite
database
🛠️ infrastructure
.github/workflows
contains a number of CI actions
- linters.yaml: Runs linting checks on the code
- release.yaml: Make and pushes docker files on a new code release
- test-docker.yaml': Runs tests on every push
The docker file is in the folder infrastructure/docker/
The version is bumped automatically for any push to main
.
Environmental Variables
- DB_URL: The database url which the forecasts will be saved too
- NWP_PATH: Override NWP data path. THis is useful when running this locally.
- FAKE: Option to make fake/dummy forecasts
- MODEL_NAME: Optional of 'nwp_simple' or 'nwp_simple_trained'
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for nowcasting_forecast-1.2.3.tar.gz
Algorithm | Hash digest | |
---|---|---|
SHA256 | 81b0fe460dda1cd968f120219e60427ad3bd8d058fa251992d33d79b6498e5c3 |
|
MD5 | bd648a2d2d287d0fde30cbb71349909a |
|
BLAKE2b-256 | 42cd4b65abfcb58cbaf77f21244199fc14ddf8f7ed6dd2236fdd2af52898fd83 |
Hashes for nowcasting_forecast-1.2.3-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | a67afd689eead4263fe962fea40a8ad68e03be54ea0c02c97b6418f3be6cd0c3 |
|
MD5 | fce9f16839152e41d9c87fbeea46be1c |
|
BLAKE2b-256 | e593ddfc4c6e4b11b148bd32c65886055de325c471d94b48ad8e69994ad8a11b |