Bachelor's thesis Felix BŁnning

Development of a Modelica-library for the Agent-based Control of HVAC Systems

Due to continuous global warming and the declining reserves of fossil fuel, an increasing number of laws promoting the integration of renewable energy sources into the current energy generation have been passed in the recent years. As the conditioning of building climate accounts for more than one third of the primary energy consumption in OECD countries, also this area underlies increasing regulations, which lead to the use of complex HVAC systems with the aim of lowering emissions and energy consumption.
The complexity of such systems leads to the need for new control strategies which allow stable and efficient operation of these systems. The approach of Multi-Agent Systems has been successfully applied in various scientific domains, handling complex control problems, and therefore promises good performance in the domain of HVAC control. For the numerical simulation of such systems, the programming language Modelica is commonly used. However, there is no library for the use of Multi-Agent Systems available in Modelica.
In the course of this work, concepts for the implementation of agent logic and agent communication protocols in the Modelica language are developed first. These concepts are validated on the basis of a book-trading problem in comparison with the Java-based agent tool JADE. Afterwards, a Modelica Library for the agent-based control of heat and cold generation in HVAC systems by using UDP communication and exergy-based cost functions is developed. The components of this library are then validated with the help of a numerical model of a complex building energy system, which includes a variety of heat and cold suppliers.
The validation shows that the developed agents present a reliable option for the control of complex building systems. Beside the abidance of comfort constraints, a 2% reduction of exergy destruction over the simulated period is accomplished compared to a reference mode-based control concept. With further improvement on the agent behaviour higher reduction is probable. However, the simulation time is increased significantly by the agent system.
It is proven that Multi-Agent Systems are suitable for the control of building energy systems. Additionally it is shown that theModelica language allows the implementation of complex agent systems including FIPA compliant communication structures. The poor computational performance of the agents presents a drawback of the concept and therefore needs further investigation. Furthermore, for the control of heat and cold distribution rather than heat and cold generation, the development of model predictive cost functions leaves roomfor further research.