Big data raises new opportunities for deep insights and supporting decision-making. To seize these opportunities, methods that derive useful knowledge from large amounts of data are needed. Such methods can help meet urgent challenges in many fields. An urgent challenge for energy systems is the necessary transformation towards sustainability to mitigate climate change. One crucial aspect of this challenge is a permanent optimal operation of energy systems. In principle, mathematical optimization can best determine the optimal operation of energy systems. However, manual model generation and operational optimization of energy systems are time-consuming and can thus prevent an application of mathematical optimization in practice. This thesis presents methods that use measured data to automatically generate mathematical models of energy systems to tackle the challenge of time-consuming model generations. Additionally, methods are presented that accelerate the operational optimization of energy systems. Regarding model generation, the presented data-driven methods solve the trade-off between accuracy and computational efficiency of the energy system model by weighting each component model by its role in the overall system. Thereby, the methods automatically generate energy system models that allow for accurate and computationally efficient optimization.
To accelerate the operational optimization of energy systems, we present two methods that decompose the complex operational optimization problem into smaller subproblems. The methods provide high-quality solutions. The first method employs expert knowledge about the individual energy system to significantly accelerate the operational optimization while retaining an excellent solution quality.
The second method applies artificial neural nets to solve the operational optimization in a reliably short time while offering a high solution quality. Overall, the methods presented in this thesis enable a broader application of mathematical optimization for energy systems.