I am a research assistant and PhD student from Technische Universität Dresden, supervised by Wolfgang Lehner and Maik Thiele. My PhD thesis is on feature-based time series analytics. It focuses on statistical and learned representations for time series datasets and their use in data-mining tasks such as modification, generation, forecasting, classification, and clustering.
The GOFLEX project integrates a group of electricity smart-grid technologies, enabling the cost-effective use of demand response in distribution grids. The intermittent generation of renewable energy sources and the fluctuating electricity consumption require accurate and efficient forecast techniques in order to safely support an increasing share of renewable electricity generation.
Technische Universität Dresden has developed the benchmark framework ECAST for energy forecasts that allows transparent comparison of forecasting models for different forecast scenarios and requirements. I integrate ECAST as part of the GOFLEX cloud service and provide additional forecast models for energy forecasting.
GOFLEX Project (Generalized Operational FLEXibility for Integrating Renewables in the Distribution Grid)
For more than three decades, researchers have been developing generation methods for time series from the weather, energy, and economic domain. My goal is to provide a comparative and cross-domain assessment of generation methods and their expressiveness. Moreover, I propose generation methods based on different recombination techniques in order to evolve time series datasets that are highly similar to the given dataset.
Time series usually represent actual data from organization and business processes. For critical tasks like decision-making, planning, predictions, and analytics, it is also desirable to apply them to alternative scenarios in order to prepare for developments that diverge from expectations or assess the robustness of current strategies. In this project, I proposed a generally applicable and easy-to-use method for creating what-if scenarios on time series datasets. The extraction of descriptive features allows the construction of an alternate version by means of filtering and modification of these features.
The FFQ project (Flash Forward Query Framework) has the goal to integrate model-based forecasting into a database management system. It offers a simple interface to make forecasting usable for any database user. Within this project different topics are addressed that aim to speed up the forecasting approach inside the database while providing a high accuracy.
I developed techniques for maintaining forecast models in the DBMS so that they keep track of the updated underlying data. Moreover, I integrated maintenance techniques for derived forecasts such as, e.g, aggregated and disaggregated forecast schemes.