Modelling of the complex data space: Architecture and use cases from NRG4CAST project

The following contribution offers technical solution for heterogeneous multivariate data streaming modelling built on top of open-source QMiner platform. The presented infrastructure is able to receive data from different sources (sensors, weather, weather and other forecast, static properties, …) with many different properties (frequency, update interval, …), it is able to merge and resample this data and build models on top of it. Technology was used to prepare prediction models for 5 different energy related use cases which include public buildings, thermal plant production, university campus buildings, EPEX energy spot market prices and total traded energy. Average relative mean absolute error of the model predictions varies between 5–10%, and qualitative analysis of predictions shows significant correlation between predictions and true values.