Motivation
It is believed that 45% of all cars sold in Norway in 2018 will be electric, and politicians have set an ambitious goal saying that in 2025, all cars sold should be zero-emission. In parallel, Statens Vegvesen announced in June 2018 that they will invest 450 MNOK into ICT technology and the development of smart roads. Both these facts signal that the transport sector is changing.
At IFE (Institute of Energy Technology, http://www.ifi.no/ ) we are developing strategies for a zero-emission transport sector in Norway. Through partnerships in on-going projects, we have recently been given access to nation-wide high-quality data on fast-charging of electric vehicles. These are being used as an aid in the design of future large-scale charging infrastructures. However, we do also see that it would be possible to use this data to predict power demands at specific locations. Such a prediction is crucial for an efficient management of the charging infrastructure, due to two reasons, firstly to enable planning of driving routes such that EV owners reduce queueing time at charging stations, and secondly to dimension and control storage capacity at the stations so that grid companies can deliver the requested power without overly expensive grid reinforcements.
Goal
The primary goal is to predict week-ahead power flows at several important locations in Norway. As a secondary goal we wish to establish a discussion on how we potentially could automate and improve the model with new types of data.
What will you do?
- Screen the data and look for five clearly different charging stations to use in the subsequent machine learning models.
- Test different machine learning models and compare the accuracy for all five charging stations.
- Establish a real prediction for the following week in order to, within the project, compare the best practice model with actual charging patterns.
- Propose a solution for automation and continuous learning of the prediction model. We’ll have more vehicles on the road every year. We’ll likely also see an increase in battery size. And eventually also an introduction of semi- and full autonomous vehicles. These three parameters will have a major impact on the charging patterns, and we need to continuously develop the model. What if you had real-time access to the database, how would you automate your training model such that it would “re-iterate” every month with partly new data?
- As the last point we wish to have suggestions for a “wish-list” of data that you think would be interesting to include in your training model. Think physical phenomena that you believe is important and suggest alternative ways to collect this data in a continuous manner.