An important task of the electricity distribution companies is to forecast the electrical load (demand) for a given sub-network of consumers, as this is a relevant for identification of critical points and making decisions to buy or sell energy. From a data mining perspective, this problem is characterized by a large number of variables (data from the electrical sensors), produced in a continuous flow in a dynamic non-stationary environment. Standard prediction techniques in such circumstances fail and more complicated dynamic models are required models that evolve over time and are able to adapt to changes in the distribution generating examples. Besides only forecasting the electricity load, the prediction should also contain an explanation of the phenomenon and the quality of the predictor must be clearly assessed with prediction reliability estimates in order to support users final decision.

In the framework of the project, the both research groups will aim to develop decision explanation methodology for concept drift detection in data stream learning, develop and test reliability estimates suitable for data streams, develop online summarization techniques for electrical time series, and study the utility of reliability estimates and decision explanations in data streams. The deliverables will ensure better predictions and robust decisions in electrical network management.