D4.2 General Draft of MIRACLE Forecasting Approach

Most renewable energy sources (RES; e.g., windmills or solar panels) pose the challenge that the production depends on external factors such as wind speed and the amount of sunlight. Hence, available power from RES cannot be planned as traditional energy sources. As a result, there is the need for balancing energy demand and supply in order to integrate more renewable energy sources. Accurate and efficient forecasts for short-term and mid-term horizons of energy consumption and production are a fundamental precondition for this dynamic and fine-grained scheduling of energy demand and supply. The state-of-the-art of forecasting energy demand and supply mainly focus on high accuracy forecasting of energy demand, while only few techniques exist for energy supply. In addition, there are further challenges. First, with regard to balancing energy demand and supply, forecasting takes place in a distributed system architecture that is inherently given by the hierarchy of involved organizations. Second, the large scale of the distributed system, in combination with a continuous stream of updates, leads to the requirement of efficient forecasting and forecast model adaptation.

In this article, we describe the general \miracle forecasting approach from a holistic perspective. This includes, (1) the requirements of forecasting as precondition for scheduling energy demand and supply, (2) an approach for synchronization of forecast models in distributed environments and their integration into the system architecture, (3) the description of the two primarily used high accuracy forecast models and their usage, and (4) techniques for efficient continuous forecast model evaluation and adaptation. Finally, our experimental evaluation investigates the accuracy of the introduced techniques.  In conclusion, this overall description of the \miracle forecasting approach is the precondition for the scheduling of energy demand and supply. This approach already shows good accuracy. However, it can be further enhanced in the future by taking into account domain-specific characteristics, external information sources and the near real-time scheduling process.

Authors:
Matthias Boehm, Ulrike Fischer, TUD; Lars Dannecker, SAP; Zoran Marinsek, INEA