Modelling in Energy Related Scenarios

Fusing heterogeneous multivariate data in stream mining scenarios is a demanding task. Successful fusion requires a wellthought approach. We propose the use of a stream processing engine (SPE) that enables implementation of all the needed methods and ensures almost real-time responsiveness of the system. In the paper we propose an infrastructure that is able to receive data from various heterogeneous sources (static properties, weather data and forecasts, other forecasts, and primarily sensor data). In the implementation of the proposed infrastructure we address issues related to the heterogeneous nature of the data, like different frequency, different update interval, and different nature of the data. The pipeline was used to prepare stream prediction models for five different energy-related use cases, which include public buildings, a thermal plant production, university campus buildings, and EPEX energy spot market prices alongside the total traded energy.