The STORM project was awarded at the DHC2016 conference, the 15th colloquium on urban heating and cooling, which was held in September in Seoul.
At the conference, Dirk Vanhoudt (VITO) and Christian Johansson (NODA) presented the project, together with their research, on an energy demand forecasting algorithm.
This research has been awarded with the “Excellence in Research on Urban Heating and Cooling” award from the International Energy Agency and the Korean District Heating & Cooling Association.
The STORM project is a collaboration between several companies, supported by the H2020 program of the European Union.
The project aims to improve energy efficiency at the network level by developing an innovative intelligent controller capable of learning by itself.
This controller will be installed on two pilot projects:
Mijnwater BV (The Netherlands)
Rottne (Sweden)
The research in question focused on energy demand forecasting algorithms that were able to learn for themselves over time.
This forecast is very important for the STORM controller, since it is the one that allows it to know when to intervene.
In the beginning of September 2016, the STORM consortium was present at the 15th International Symposium on District Heating and Cooling in Seoul, South Korea. Dirk Vanhoudt (EnergyVille/VITO) gave a general presentation of the STORM project and the latest status. Moreover, Christian Johansson (NODA) presented a research paper on the first results of the energy demand forecasting algorithm, which is part of the STORM controller. We are proud to announce that this paper won the ‘Award for Research Excellence in District Heating and Cooling’ by the International Energy Agency’s District Heating and Cooling Programme (IEA DHC) and the Korean District Heating & Cooling Association (KDHC).
Dirk Vanhoudt at DHC2016
Dirk Vanhoudt presents the STORM project at DHC2016
What is the paper about?
The awarded paper describes a number of self-learning algorithms to forecast the energy consumption of a heating or cooling network for the next 24 hours. Indeed, this forecast is of high importance for the STORM controller because it identifies the moments where the controller must intervene.
The forecast is based on two types of machine learning algorithms to forecast the future consumption in the network, namely Extra-Trees Regressors (ETR) and Extreme Learning Machines (ELM). Both algorithms use the historic consumption data in the network to ‘learn’ the relationship between historical data and the forecasted future consumption. More specifically, the first method uses tree-based algorithms and the second uses a type of neural networks.
These algorithms have already been implemented in the demonstration network in Rottne, Sweden. Hence, the paper presents the results of three months of online heat load forecasting, i.e. January to March 2016. All in all, the algorithms need about three weeks to learn the network behaviour and to be able to produce sufficiently good results. In general, the performance of the ELM algorithms is slightly better than the ELM algorithms. During February, the prediction was best, leading to very good ‘mean absolute percentage errors’ of respectively 7.6 and 6.8%.
Read the whole paper here: Operational demand forecasting in district heating systems using ensembles of machine learning algorithms
Christian Johansson receives DHC2016 award
Christian Johansson from NODA receives the award for research excellence on the STORM controller
Next steps
The STORM consortium sees this award as motivation for their future research efforts and to continue improving the controller.
