Florian Scholz

Abstract

In the first amendment to the Federal Climate Change Act, the year 2045 was set as deadline for achieving net greenhouse gas neutrality in Germany. An important building block in attaining this goal are integrated renewable electrical energy systems. The combination of different power generation and storage technologies as well as participation in electricity trading on spot markets ensures that the high flexibility requirements resulting from the lack of base load capacity of renewable electricity generation can be met. The economically optimised control of such energy systems poses considerable challenges for system operators. They are confronted with the stochasticity of load profiles and the limited controllability of renewable power plants. Moreover, the pronounced volatility of market prices in intraday trading and very short lead times in the provision of electrical power make high demands on the real-time capabilities of the system control.

As part of this PhD project, we develop heuristic control strategies that can be used, for example, by regional power suppliers or energy communities. The system operator deploys a network of various controllable and non-controllable electricity generators such as wind turbines, photovoltaic devices, or fuel cells, electricity storages, storage facilities for green fuels like biogas or green hydrogen, and facilities for the production of energy carriers such as electrolysers. The firm operating the system can act as a supplier or consumer on the electricity market. It is obligated to cover the electricity requirements of his customers, which include private households and RLM customers like small and medium-sized enterprises. Surpluses and deficits in power generation can be balanced both by means of its own storage facilities and on the electricity market. The electricity market is represented as commodity exchange with continuous trading of 15-minutes intraday products.

The control problem is formulated as an average-reward Markovian decision problem. A linear program based on a deterministic ex-post consideration of the problem serves as a benchmark. The objective of our research consists in developing parametric stationary control policies, which can be described by a few parameters and are easily deployable. The optimisation of the control parameters is typically based on monotonicity properties of the reward function or the transition law of the Markovian decision process.