Starting point / motivation

Increasing digitalization and networking in building technology is making the operation of buildings multi­dimensional and complex, which is leading to growing demands for intelligent control and regulation systems.

Traditional control systems are reaching their limits with increasingly complex dynamic systems such as heating, air conditioning, and ventilation (HVAC) systems in large buildings—especially when predictive approaches are needed to respond proactively to changing load profiles, malfunctions, or environmental conditions, which reduces the energy efficiency of buildings.

Scalable and replicable solutions are needed to exploit existing efficiency potential in new and existing buildings. In recent years, AI and cloud-based approaches have been increasingly used. These are usually costly to integrate into buildings and contribute to rising electricity demand for computing power due to the computing effort required.

Contents and goals

To increase energy efficiency, we propose the use of edge AI in conjunction with specialized hardware in the form of neuromorphic chips.

The project has the following three objectives:

1. To develop a scalable control technology solution "on the edge" to increase energy efficiency and extend the service life of heating, air conditioning, and ventilation systems.
2. Investigation of the use of hybrid learning methods and neuromorphic computing technologies for PID parameter optimization, adaptive control, and pattern recognition of control anomalies via spiking neural networks.
3. Proof of concept of TinyML models, metaheuristic algorithms, and spiking neural networks on edge devices on a laboratory scale for selected systems.

Methods

The following methodological approach was chosen to achieve the project objectives:

1. Creation of a requirements definition and evaluation criteria, as well as use cases to be processed.
2. Implementation of the optimization and diagnosis of building systems. In doing so, different methods and procedures of machine learning and model-based diagnosis are to be implemented and their results compared with each other.
3. Evaluation of the methods using the same use cases and evaluation criteria.
4. Combination of the best methods for optimization and diagnosis, and in the final step
5. Initial evaluation of the combined method using the use cases in a practical setup, including an assessment of overall performance.

Expected results

Implementation of modern AI methods to improve the control and diagnosis of building systems. The methods are to be executed on current edge hardware and neuromorphic chips and increase the overall efficiency of buildings.

The goal is to improve control quality by at least 60% compared to state-of-the-art autotuning methods. In addition, we expect scientific publications in high-impact journals and conferences summarizing the proposed solutions and their results.

Project management

Technische Universität Graz - Institute of Software Engineering and Artificial Intelligence

Project or cooperation partners

• Forschung Burgenland
• DiLT Analytics FlexCo
• Engynear tga GmbH

Graz University of Technology
Institute of Software Engineering and Artificial Intelligence
Prof. Dr. Franz Wotawa
Inffeldgasse 16b/2
A-8010 Graz
Tel.: +43 (316) 873 5724
E-mail: wotawa@tugraz.at
Web:www.tugraz.at/institute/sai/home