- We develop a system to identify faults of heat pumps from smart meter data. 

- We enhance the smart meter data with data about local weather conditions and geo locations. 

- We cooperate with companies in the energy service domain to make the research go live in real-world applications.

Heat pumps are already installed in around 20% of buildings in Switzerland. This development reduces the use of fossil fuels, but increases the demand for electrical energy. Accordingly, efficient operation of the units in practice is important even many years after installation. However, for installers, the design and configuration of heat pumps is a complex task, and often characteristic curves are set conservatively and complaint-free systems are not optimised further. Households are usually unable to assess whether the electricity consumption of their system is appropriate or whether there is potential for savings. 

- Our system is designed under data minimization constraints, such that sensible household characteristics are undisclosed. 

- We investigate how reliable predictions can be achieved even with only small data sets. 

- We target to derive a set of regular patterns of heat pumpt faults, alongside predictions of cost savings if fixed. 

- We aim at systems that continue learning and can be improved while running live. 

Harvesting Open Data for Household Classification    

Testing the Transferability of Household Classifiers

Identifying Target Customers for Cross Selling: The Case of Biogas