Update 01 July 2020: This repository provides the code to generate your own variations on a seed file. For commercial purchase of projected files for simulation, based on climate model outputs, please email info@arbnco.com to request a quote.
This repository contains scripts to create synthetic weather time series from a short weather record of at least one year. This tool is named indra (1). All the scripts here should be treated as experimental unless explicitly stated otherwise.
You need Python 3.14 and uv. From the repository root:
uv sync
Tell indra where your weather data lives. Copy .env_example to .env and set INDRA_WEATHER_DIR to the folder holding Historical/ (one sub-folder of EPW files per station) and, for climate-change runs, CMIP6/ (one sub-folder of daily projections per city). .env is not tracked, so each machine keeps its own.
Say what to generate in config.toml. The [run] table sets the defaults: how many synthetic years (n_samples), the input format, the ARMA model orders, and whether to apply a climate-change scenario (climate_change, cc_scenario, epoch). Each [[stations]] block names a station, where its outputs go, and its input paths, written as ${INDRA_WEATHER_DIR}/.... Relative paths are taken from the folder the config sits in.
Run it:
uv run --env-file .env indra --config config.toml
indra learns a model from each station's record, then writes one synthetic weather file per sample into that station's store_path. uv run indra --help lists the options for a single run without a config file.
Tests: uv run --env-file .env pytest. The end-to-end test trains on the Atlanta record and writes one synthetic year, which takes about five minutes, so it only runs with INDRA_RUN_INTEGRATION_TESTS=1 set.
The package lives in src/indra/. lib/ holds the legacy example (uv run python lib/vali.py) and the old installers.
If you're interested in reading the methods used first, see the list of papers given below.
The original MATLAB/R scripts implement the algorithms published in Parag's thesis. They are documented in the two conference papers and the thesis below, but they are not in this repository and I won't be working on them any more. The Python package here is a translation of them. The method is an almost-completely-faithful translation of the MATLAB scripts(2).
This work derives from my PhD thesis at the Ecole Polytechnique Federale de Lausanne, EPFL (Chapter 3). The method is described in the following publications/references:
- P. Rastogi and M. Andersen, ‘Incorporating Climate Change Predictions in the Analysis of Weather-Based Uncertainty’, presented at the ASHRAE and IBPSA-USA Building Performance Modeling Conference, Salt Lake City, UT, USA, Aug. 2016, [Online]. Available: https://infoscience.epfl.ch/record/218596.
@inproceedings{rastogi2016incorporating, title={Incorporating Climate Change Predictions in the Analysis of Weather-Based Uncertainty}, author={Rastogi, Parag and Andersen, Marilyne}, booktitle={Sixth National Conference of IBPSA-USA, Salt Lake City, UT}, year={2016} }
- P. Rastogi, ‘On the sensitivity of buildings to climate: the interaction of weather and building envelopes in determining future building energy consumption’, PhD, Ecole polytechnique fédérale de Lausanne, Lausanne, Switzerland, 2016. Available: https://infoscience.epfl.ch/record/220971?ln=en. DOI: http://dx.doi.org/10.5075/epfl-thesis-6881.
@phdthesis{rastogi2016sensitivity, title={On the sensitivity of buildings to climate: the interaction of weather and building envelopes in determining future building energy consumption}, author={Rastogi, Parag}, year={2016}, school={Ecole Polytechnique F{'e}d{'e}rale de Lausanne} }
- P. Rastogi and M. Andersen, ‘Embedding Stochasticity in Building Simulation Through Synthetic Weather Files’, presented at the 14th International Conference of the International Building Performance Simulation Association, Hyderabad, India, Dec. 2015, [Online]. Available: http://infoscience.epfl.ch/record/208743.
@article{rastogi2015embedding, title={Embedding stochasticity in building simulation through synthetic weather files}, author={Rastogi, Parag and Andersen, Marilyne}, journal={Proceedings of BS}, year={2015} }
This tool is distributed under the GPLv3 license. Please read what this means here.
The Python package needs only free software. While you are free to use the scripts as you please, I am not liable for anything that happens as a result of using my scripts. Like if you accidentally release nuclear missiles, ruin the ski season in Switzerland, or cause a drought in Scotland.
Please cite this work using the citation for my thesis:
Rastogi, Parag. 2016. ‘On the Sensitivity of Buildings to Climate: The Interaction of Weather and Building Envelopes in Determining Future Building Energy Consumption’. PhD, Lausanne, Switzerland: Ecole polytechnique fédérale de Lausanne. EPFL Infoscience. https://infoscience.epfl.ch/record/220971?ln=en.
@phdthesis{rastogi_sensitivity_2016, address = {Lausanne, Switzerland}, type = {{PhD}}, title = {On the sensitivity of buildings to climate: the interaction of weather and building envelopes in determining future building energy consumption}, shorttitle = {Sensitivity of {Buildings} to {Climate}}, url = {https://infoscience.epfl.ch/record/220971?ln=en}, language = {EN}, school = {Ecole polytechnique fédérale de Lausanne}, author = {Rastogi, Parag}, month = aug, year = {2016}, note = {doi:10.5075/epfl-thesis-6881} }
If you have questions or concerns, or notice errors, please contact me at contact[at]paragrastogi.com.
Happy creating fake weather!
-
I haven't thought up a smarmy, contrived acronym yet but I'm working on it.
-
The difference lies in the 'simulation' of the SARMA model to produce synthetic 'de-mean-ed' series. I am not convinced that I should reproduce the 'custom noise' functionality used in the old scripts to simulate the SARMA models with bootstrapped residuals. For now, I am doing a 'conventional' simulation by using white noise.
- This work began during my PhD at the Ecole Polytechnique Federale de Lausanne. It was funded by the CCEM SECURE project and the EuroTech consortium. February 2012 - August 2016 2.Subsequent work and the Python translation was carried out when I was a visiting scientist at the Energy Systems Research Unit (ESRU), University of Strathclyde, Glasgow, and the RIKEN Institute for Advanced Intelligence Project (RIKEN-AIP), Tokyo. My stay at these institutions was financed by the Swiss National Science Foundation (SNSF) under grant number p2elp2_168519.
I woud like to thank my hosts: Prof. Joe Clarke (Strathclyde, Glasgow) and Dr Mohammad Emtiyaz Khan and Prof. Masashi Sugiyama (RIKEN-AIP, Tokyo). The advice of Prof. Anthony Davison (EPFL, Lausanne) was crucial in creating the first models.
Parag is the sole author of these scripts. The original PhD thesis was supervised by Professor Marilyne Andersen. The scripts are the intellectual property of Parag and EPFL. Their redistribution under the liberal GPLv3 license is with the approval of EPFL, acting through Prof. Marilyne Andersen.