23.04.2018 | Three days of programming in a row - and what comes out of it

On the keys, ready, go!

They are programming for three days at the foot of the Limmern pumped storage plant. ETH students are working with power trading specialists from Axpo to work on new forecasting models for the energy business. What is a hackathon anyway? And what does the occasion have to do with machine learning? Here it is explained.

Axpo's Martine Graziano, Head of Governance & Portfolio Management, sees great potential in Hackathon: know-how can simply be exchanged between internal and external specialists and further questions can be tackled. How she assesses the nature of the collaboration:

Martine, where did the idea to run such a Hackathon with students come from?
The idea to run a Hackathon evolved during a digital innovation workshop in Madrid where we discussed on how forecasting know-how can be leveraged and effectively improved across the different units of Axpo Trading. As such, the hackathon was part of the initiative "Forecasting Lab". In order to foster an agile approach, we initially ran an internal Hackathon in February to address a Spanish forecasting challenge using machine learning with 28 participants from various business divisions and hubs. This hackathon was also meant as prototype and the corresponding learnings were applied to the running of the three-day Hackathon in Tierfehd. The idea of running both an internal and an external hackathon was also to be able to better judge the value of having students with new approaches and ideas participating in such an event.

What was the goal of the Hackathon and what results were produced?
The challenge to the six teams, each consisting of one Axpo employee and two students, was to predict with machine learning the power imbalance position of the Swiss control area. The topic is similar to the one from the first hackathon and allows us to compare the slightly different problem challenge. The teams managed within a very short timeframe to test various approaches concerning their relevance in terms of which data to use for training/running the model as well as of which model approach to use. The developed models and gained experiences are currently being processed and structured in order to then share and discuss them with forecasting experts from SEM, Nordics, Spain and CMT to finally enhance them until they can be used in daily operations.
In addition, we applied implicitly the principle of "prototype – measure – scale" to improve the approach and at the same time gain experience in this regard.

Was the Hackathon as success? Will there be a similar event sometimes in the future?
I think we can dare to consider the Hackathon was a great success. Besides the various technical insights and learnings, Axpo employees gained within a short timeframe new knowledge in the domain of machine learning approaches currently taught at universities. Furthermore, we managed to represent Axpo as an innovative, international and dynamic company. All participating students would recommend to their colleagues to attend an Axpo Hackathon and four participants have already applied to the Axpo Trainee program or for an internship. We are discussing to what extent and on what topic we could run future Axpo Hackathons.

And what was the first price?
The three winners have each received a LORA-Starter Kit (cf. article on the right). This kit contains a programmable micro processor and some sensors that use the low-power GSM network for data communication. LORA-sensors offer a large potential to measure various parameters at very low cost. Maybe they will be useful for parts of our business to measure inputs for forecasts – accordingly, the winners have been asked to get back to us once they successfully managed to the get the kit up and running…

More articles for you

Show all

People

Jobs for the future

Ambition 2030: 600+ apprentices

Read more

Energy market

French vegetable farmers rely on Axpo’s expertise

Growing business

Read more

Energy market

Navigating structural tightness and growing uncertainty as winter approaches

European Energy Markets Monthly, November 2024

Read more