ProbCast is an R package under continuous development by researchers at the University of Strathclyde. It is a collection of functions for probabilistic forecasting (mainly wrappers for qunatile and semi-parametric regression model fitting functions), cross-validation, evaluation and visualisation. Central to ProbCast is the data class MultiQR, for storing the results of multiple quantile regression, and methods for working with MultiQR objects.
You can read more about the first beta release of ProbCast here, and keep up-to-date on GitHub.
Control REACT: the outputs from this innovation project (2019-2021, with NGESO) include an extensive dataset of wind, solar and net-demand data for Great Britain, and code implementing state-of-the-art probabilistic forecasting methods using ProbCast. Available on Zenodo: Data & Code
ECMWF: Historic weather forecasts from ECMWF are now CC-BY and accessible via the MARS archive (free for academics in member countries)! Many datasets are also freely available via web interface or API.
GB Electricity Data: Metered data from all transmission connected and some embedded generators is available for research at 30-minute resolution from approximately 2009 to present with affiliation to UKERC or purchase. Quite a lot can be obtained for free from Elexon.
Monash Time Series Repository: includes many time series including energy demand and wind power generation at high temporal resolution, details here.
SERL: The Smart Energy Research Lab is collecting smart meter data from thousands of GB consumers as well as survey responses from participating households.
MIDAS: I have worked extensively with the MIDAS dataset provided the British Atmospheric Data Centre which provides meteorological measurements from weather stations around the UK. This data is free to use for academic purposes. If you would like to use a set I have compiled, please contact me.
HYDRA: The other on-shore meteorological dataset I have used most is the HYDRA set of potential wind across the Netherlands, provided by the Royal Netherlands Meteorological Institute. Again, if you would like a specific set which I have used in my publications, please contact me.
FINO: The FINO project comprises three offshore platforms collecting atmospheric and oceanographic data, including wind speed and wave height measurements.
Marine Data Exchange: Data relating to UK offshore wind projects. Highlights include 30 year wind resource model outputs produced by the UK Met Office.
Wiki-Energy: A smart grid project in the US including domestic solar generation and electricity usage as recorded by smart meters installed in hundreds of homes.
Historic UK Met Office Forecasts: I have not used this resource but it could be very useful if you are interested in using NWP outputs.
Global Energy Forecasting Competitions 2012 & 2014: Data from the GEFCom2012 and GEFCom2014 is quality controlled and lots of state-of-the-art competitors to test yourself against.
Mesonet: Wind and other meteorological measurements from West Texas and New Mexico.
EMHIRES: Wind power time series for European countries based on 30 years of reanalysis.
I release code as supplementary material with paper wherever possible. Here are some additional bits of code not linked to papers:
Slides and Examples from EES-UETP 2018 Lecture
"Statistical Learning for Uncertainty Forecasting." Download R code, data and slides here.
Winning Entry to EEM Wind Power Forecasting Competition
The R code written used by Ciaran Gilbet and myself for the EEM2017 forecasting cometition may be accessed here. At the present time, wind power data may not be shared due to legal restrictions but we are working with the data owners to change this.
University of Strathclyde LaTeX PhD Thesis Template
Here is a template for a Strath PhD thesis with the required formatting and content (title page, declaration etc). Please feel free to send feedback/updated code with additional features. This is also available on Overleaf.
"Very-short-term Probabilistic Wind Power Forecasts by Sparse Vector Autoregression"
In this work I used a data set comprising 5 minute resolution power measurements from 22 wind farms in south eastern Australia from 2011-2013. This data is provided by the Australian Electricity Market Operator and is public access. Please credit myself and Stefanos Delikaraoglou for its pre-processing. The data can be downloaded in .csv format here.
Code for an R function which fits a sparse VAR model can be downloaded and used under a BSD licence. It is accompanied by a basic example of how to implement it using the AEMO dataset available above. Download. *I hope to produce a more flexible version of this function soon to make it a little more user-friendly and hopefully speed it up a bit too.