Competition

Western Power Distribution Data Challenge (Part 1)

Organized by wpd_challenges - Current server time: May 2, 2025, 9:58 a.m. UTC

First phase

First phase
Nov. 9, 2021, midnight UTC

End

Competition Ends
Dec. 10, 2021, 11:59 p.m. UTC

Challenge Overview

This is the first of three Western Power Distribution (WPD) short data challenges! The aims of these challenges include:

  1. Demonstrating the value in making data openly available
  2. Increasing the visibility of some of the challenges network operators face
  3. Increase the understanding of the different ways to tackle some of these problems
  4. Providing high quality and accurate benchmarks with which to enable innovation and research

Note all the slides from the kick-off and other information can be found on our LinkedIn group 

 

Challenge 1: High Resolution Peak Estimation

This initial challenge aims to understand how accurately high resolution features can be estimated given only information from lower resolution data. Specifically we are asking participants to estimate the highest peak value and lowest trough at a one minute resolution within each half hourly period given only half hourly measurements. This is an interesting problem to a distribution network operator as the spikes in demand can mean strain on their network. Such issues may become increasingly common, especially on the lower voltages of the network, due to the expanding use of lower carbon technologies such as electric vehicles, and heat pumps. However, monitoring can be expensive (especially in the long term) as it requires investment in additional storage, communications equipment and processing units.

Problem Setup

The scenario presented by this challenge is whether limited higher resolution monitoring can be used to estimate future high resolution demand spikes when the high resolution is no longer available. Participants are asked to estimate the largest and smallest demand at one minute for each half hour over the test period (a month) given the lower resolution demand data (which is at half hourly resolution), and accompanying weather data. To develop and train the models about 2 years of historical data have also been provided and a validation phase (the month prior to the test period) has also been provided which also serves to test the submission process. In addition the actual historical high resolution data will also be available for the dates prior to the validation/test periods. The evaluation metric is a skill score which measures the difference between the actual value and your estimated value, scaled according to a benchmark model (see Evaluation page for more details).

Validation Phase

To help participants get used to the submission process and better understand the problem there will be a validation phase which runs for a two week period from 11th November 2021 to 25th November 2021 (midnight). This phase allows unlimited submissions and can help you test and refine your models. For the Validation phase teams are asked to estimate the maximum and minimum at one minute resolution for every half hour of the month of August 2021. In other words the aim is to estimate a total of 2976 values (2 values per half hour for each day (31) of the month). Results from this phase will not be used to determine your overall ranking for the challenge.

Testing Phase

The testing phase will run from 26th November 2021 to 10th December 2021 (midnight). You will be asked to estimate the same values for September 2021 (note that due to data limitations the hours after 10am on 30th September are not included). During this stage you will be limited to six submissions. This phase will be used to judge your success and ranking for the challenge.

Teams

We are allowing teams of up to 5 individuals. To group yourself in a team you need to follow the codalab guidance found here. Note that all team members can make submissions during Phase 1, but we ask that only a single nominated team member make submissions during Phase 2.

Data

The substation data is hosted on the WPD data portal and can be found here. MERRA-2 weather data from locations close to the substations can be found as public data under the participate then files tab. More details of the data are in the data tab. The license terms for usage of the data can be found here and here.

How to Submit

The data must submitted via a zip file of a CSV file with your maximum and minimum predictions. Templates for your submission can be found here. The CSV must be renamed "Predictions" before zipping in order to be evaluated. 

Prizes

The winners of this challenge will receive credit worth £500 for a cloud computing service (e.g. AWS). The top three teams/individuals from Phase 2 will be invited to present their methodologies in a webinar.

Evaluation

This section explains how the challenge will be evaluated. The challenge is split into two phases. The first phase is a validation period and doesn't count towards the final score. The second phase is the scored testing phase. The validation period will consist of predicting the minimum and maximum demand at one minute resolution for each half hour of August 2021. The first phase allows unlimited submissions since it is to help teams become familar with the platform and submission process and test their models. The testing period will require predicting of the minimum and maximum one minute resolution demand for each half hour of September 2021. For this phase only 6 submissions will be allowed, so use them wisely. The limit on the number of submissions is to reduce good scores being developed from trial and error.

Evaluation Function

The submission will be scored based on a skill score which will be the ratio of the RMSE error of your submission and the RMSE of the benchmark model (see below for the benchmark model). The division by a common benchmark ensures that the scores are all similarly scaled and can be compared with each other.

Benchmark

To validate your method we will include a simple benchmark. This will have a skill score of 1 as we are dividing the benchmark's RMSE by itself. The benchmark is to use the half hourly values provided as part of the challege for both minimum and maximum prediction.

FAQs

Q: Can we use observed data in the future (e.g. half hour power at 3:00 when predicting for 2:30) when predicting?

A: Yes. However, be aware that you won't have the data to do this for the last half hour in each month

 

Q: I'm getting a 'File not found' error when doing my submissions

A: Make sure your submission is called predictions.csv (all lower case) and is zipped up. 

Tips for Submissions

The following are some tips that will help you make successful submissions. 

The submission templates found here should be used for making any submissions. Once updated these need to be renamed to predictions.csv and then compressed as a zip file (the name of the zip file does not matter). The submission can then be uploaded on the Submit/View Results tab.

These tips may be updated with any further advice throughout the challenge.

 

The following are the rules of participation (These must be fulfilled to be eligible for the prizes):

1. Form teams of up to 5. You must add the same team name.

2. Only a single member of a team may submit solutions during the testing phase. Any members may submit during the validation phase.

3. For this Challenge you may only use data provided and no external data sets. 

4. At the end of the competition send a short outline of your methods to wpd-challenges@es.catapult.org.uk

First phase

Start: Nov. 9, 2021, midnight

Second phase

Start: Nov. 26, 2021, midnight

Competition Ends

Dec. 10, 2021, 11:59 p.m.

You must be logged in to participate in competitions.

Sign In