PhD - AI-Driven Optimization for HVAC Systems Using Knowledge Graphs­­,­­ LLMs and Neuro-Symbolic AI

Buildings account for about 30% of global energy consumption, with a significant portion used for heating, ventilation and air conditioning (HVAC) systems to ensure the thermal comfort of occupants. Most HVAC systems are controlled by rule-based methods, which often lack the dynamic flexibility needed for energy-efficient operation. To address this, Model Predictive Control (MPC) has emerged as a promising alternative but requires detailed system models that are labor-intensive to develop for each unique building. This PhD project aims to develop methods to automate and assist in the creation of optimization tools for HVAC systems in commercial as well as industrial buildings. The focus will be on integrating AI methods such as neural networks, generative AI and neuro-symbolic AI to enhance the efficiency of HVAC systems while ensuring occupant comfort. The focus lies on the conseption of tools for the semi-automatic creation of models for the optimization of HVAC systems by encoding expert knowledge into symbolic representations, which capture all relevant data points and their relationships. These data points are identified from various sources such as free-text descriptions, datapoint lists and system codes. This information is then structured into knowledge graphs or code snippets for further processing. To enable semantic understanding of the system, AI methods - including large language models (LLMs), large multi-model models (LMMs) and neuro-symbolic AI - will be evaluated for their ability to interpret natural language inputs and reason using knowledge extracted from these graphs. The resulting knowledge graph will represent the building's subsystem, making it accessible for optimization algorithms and decision models aimed at improving HVAC performance. Become a part of our team and conduct pioneering research on AI-driven methods for optimizing HVAC systems using a combination of large language models, large multi-modal-models, knowledge graphs and neuro-symbolic AI approaches. You develop tools that support the semi-automated creation of optimization models using unstructured input data (csv-files, time-series-data, P&I-Ds, etc.) as well as expert knowledge. The integration of AI techniques to automatically identify and map relevant data points for HVAC system control, utilizing generative AI as well as knowledge graph technology for data extraction and reasoning is also part of your work. Furthermore, you implement neuro-symbolic AI approaches that combine the strengths of neural networks as well as symbolic reasoning to improve system optimization and decision-making. Also, you collaborate with domain experts in AI and building automation to ensure the solutions address real-world challenges in commercial as well as industrial settings. You will publish your research in leading academic journals and present your findings at top-tier international conferences. Finally, you ensure that your research contributes to the development of more energy-efficient and intelligent building technology systems.