Optimizing energy demand, indoor environment, and robustness of buildings using generic room types and ma-chine learning
When designing buildings, the combinations are endless. Therefore, designs are most often developed based on experience and rule of thumbs. By developing metamodels to characterise the most important rooms in terms of energy use and indoor environmental parameters the aim of this project is to enable evidence-based building design throughout the entire design process. Metamodel’s also provide the possibility of taking variability of input such as time of use, number of people and plug loads into consideration in the design face, optimizing not only according to building code but also to the expected use of the building through its lifetime.
In early stage building design, a large range of important design parameters, shape of build-ing, façade-layout and HVAC systems are determined. These are essential for energy use, overall performance and prize of the completed building. Building performance simulation (BPS) programs require hundreds of detailed inputs to evaluate one building design at a time. This approach fits poorly with the information level and uncertainties in early design. Therefore, early stage building design is often based on experience and rules-of-thumb that due to the rapid changes of energy and sustainability requirements quickly become obsolete. The aim of this project is to make simulation-based evaluation of energy demand and indoor environment easier accessible to the design team and enable exhaustive design exploration encompassing the countless combinations in early design. Due to the low level of detail and large uncertainties, it is possible to, based on design briefs, industry guidance and expert in-ter-views, to setup generic room models. For these rooms, Monte Carlo simulations are used to vary all significant design parameters and systems together with weather data and user behavior. As the number of combinations are countless, the performed simulations are used to train fast accurate metamodels using Machine Learning, basing the decision support in the early design on the results of the derived metamodels. The metamodels of all the generic room types are validated to a sufficient accuracy by comparing to validated BPS programs. Finally, the developed generic metamodels are tested in an integrated design process of both new and retrofit buildings.
In this project, co-finansed by ELFORSK, the use of generic rooms has been investigated as a replacement for rooms with complex geometries, such that meta-models can be trained based on the generic models. The investigation of the generic rooms showed that they can replace models with complex geometries, with a minimal reduction in precision.
As the generic rooms showed a high similarity with the complex geometries, there has further been investigated the usage of machine learning to train meta-models, which can be used instead of actual simulations. This makes it possible to make informed decisions, that have an impact on the energy use and indoor environment, early in the design phase and throughout the project. These metamodels have shown great potential, as they can, with high precision, calculate results for the indoor environment and energy use.
When testing the tool, it has been shown that the requirements for the indoor environment can be fulfilled, and at the same time achieve operational energy use that is half the requirement from the Building Code.
The tool has been tested on specific cases, where the applicability and the value have been highlighted. Lastly, the handling of robustness has been investigated, supplemented with sensitivity analysis.
Key figures
Category
Participants
| Partner | Subsidy | Auto financing |
|---|---|---|
| Aalborg Universitet (Fredrik Bajers Vej) |
