Energy data often consists of a growing sequence of sensor values without a designated end. To design and employ smart energy systems it is important to continuously analyze such data without slowing down due to the amount of processed data. In order to enable such analytics, it is necessary to have efficient tools available. This doctoral thesis focuses on the task of determining (in)dependencies, quantified as Mutual Information, in a data stream. Monitoring such dependencies over time can give insights about the underlying system. For instance, if two sensor attributes become less dependent, it could imply a change or degradation of the system producing these values. Due to rising speed and volume in data acquisition, a special focus of the developed methods is their scalability in regards to the number of data points. To strive for optimal solutions, the computational complexity of a method is compared to the theoretical minimum for its task.
Vollmer, M.; Rutter, I.; Böhm, K.
2018. In: 21st International Conference on Extending Database Technology (EDBT'17), pp 49-60. http://dx.doi.org/10.5441/002/edbt.2018.06