Shaping your Energy Future with VoltKraft : ConnHack 2018
ConnHack 2018 — Shape the future of Connected Life
From 9th to 11th of March 2018, Connected Living and its partners SAP, Innogy, Vattenfall, Conrad Connect, DAI — Labor from TU Berlin, IOLITE, Fresh Energy and Signal Cruncher jointly organized their first hackathon at SAP premises in Berlin. More than 50 developers participated in the event. They were provided with huge amount of anonymized data ranging from few hundred megabytes up to a Terabyte of data (from Fresh Energy) by the participating companies from the fields of Smart Home, energy and IoT (Internet of Things). The idea of the hackathon was to develop new data based services, use cases, and monetization approaches from a real data set.
If there is one thing, the hackathon differed from most other hackathons I have been is the idea of design thinking approach. During the entire hackathon, the teams were supported by mentors and design thinkers. Before the final presentation, there were two interim presentations on the penultimate day, where each team had the opportunity to pitch their ideas. These two interim presentations carried 10% of the total evaluation each and are evaluated by participants from other teams (i.e., one cannot vote for their own team). Final pitch amounts to 80% of the total evaluation and are awarded by the organizers of the hackathon which are in turn awarded based on creativity (20%), business case or potential of commercialization (40%) and implementation of the prototype (40%).
Power Consumption Data from Fresh Energy
For the challenge, we used data provided by Fresh Energy. It is a startup based in Berlin. They provide free smart meters to interested households to monitor their energy consumption for every 2 second time interval. In return, consumers get real time updates of the power consumed by the electrical appliances in their house for every day or for the whole month. This enables an individual to keep track of his/her power consumption and optimize thus reducing costs.
Anonymized data were provided from 1000 different meter-IDs and each meter-ID had one year of power consumption data for all the three phases and at 2 second time interval. So, it is about 15 million data points per meter-ID per year corresponding to 1 GB per meter-ID. All together, we were given data from 1000 different meter-IDs and the whole data amounted to about 1 TB of power consumption data. Considering that the dataset is huge and the possibility to explore such a large dataset interested our team and we decided to go with their datasets.
If there is one thing, the hackathon differed from most other hackathons I have been is the idea of design thinking approach.
Our model architecture is shown below in the figure. The whole dataset was provide through SAP cloud platform. Popular machine learning libraries such as sklearn for clustering algorithms and Keras with Tensorflow backend was used for deep learning models. We decided to use Cassandra/Neo4j as a database management and Kafka connect for connecting to external databases. Flask was used to build restful APIs and communicate with the web frontend.
Our idea was to look for potential clusters from the given set of 1000 meter-IDs, so that we could predict the power consumption pattern of each cluster and give them with relevant recommendations or offers. A typical cluster analysis for energy usage would be like the following figure. Depending on the power consumption, we could easily cluster easily households and commercial establishments. Even between households, one can cluster between small families and bigger families and have different offers based on the usage pattern.
Next, we wanted to forecast the total energy consumption in a grid. The idea behind that is to predict the supply vs demand mismatch faced by the grid operators. Power production from renewable energy sources such as wind, solar varies a lot during the day or even around the year. Sometimes, there is a surplus production of energy while sometimes there is a deficit in energy production with respect to the demand. In these cases of energy deficit, the grid providers have to switch on a backup power generator to be able to compensate for the sudden surge in power consumption.
All together, we were given data from 1000 different meter-IDs and the whole data amounted to about 1 TB of power consumption data.
Having an standby power generator just to provide power during the sudden surge in consumption is costly to maintain. To avoid this situation, the grid providers can send messages to consumers to use less power for a certain duration, when the demand is more than the supply and in exchange they will be given some incentives or benefits. For this we decided to offer the consumers with incentives in the form of cryptocurrency called VoltCoins. These VoltCoins can be exchanged for money or as gift coupons or as payment for the next month’s electricity bill.
Considering that this is an hackathon and we are given hardly 48 hours to come up with a prototype, we decided not to use the full 1 TB of data for time series forecasting as it is a computationally intensive task. We decided to predict the pattern for one user as a proof-of-concept and the graph below shows the real data vs the predicted pattern.
We used a long short term memory network to predict the total power consumption pattern on a data aggregated for every one hour for the whole year. The total power usage can be forecasted and in the case of an anomaly in the power consumption pattern, the corresponding individual can be informed/warned by a sms message to cross check his energy spendings or to look for possible defects in their electrical appliances in case if it consumes more energy.
Conclusion
In conclusion, our team came up with a solution to predict and forecast the energy production and consumption pattern. In case of mismatch between the supply and demand, Grid providers can send messages directly to the consumers to reduce their power consumption and in return give them incentives in the form of VoltCoins which can be used in exchange for money or as a gift coupons or to deduct electricity bills for the following month.
Team behind VoltKraft
Dr. Karthick Perumal from the Porsche Digital Lab, along with Marcus Jones, Dr. Juanjiangmeng Du, Dr. Setareh Sadjadi and Tiago Oliveira, all from the current batch of Data Science Retreat, Berlin participated as a team in this hackathon and won the first prize.
Karthick Perumal is a data science consultant at Porsche Digital Labs. You can follow him on Twitter and on LinkedIn.