AI for Performance
AI for Performance
Details
Energy simulation and performance prediction
During my time at TU Berlin, I took “AI for Performance” from the Department of Digital Architecture and Sustainability, TU Berlin with Prof. Dr.-Ing. Philipp Geyer.
To create a sustainable built environment, designers and planners need instant feedback on building performance, such as energy demand, comfort, or environmental impact. This is what Machine learning is exploring in one aspect. In the course, we developed an AI component for quick prediction and performance improvement.
We worked in groups and we were exposed to the advantages of Machine learning in Architecture and Urban Design.
My group worked on glassing energy performance. We explored different glass types in an attempt to evaluate building energy performance and in the end determine the best-optimized glassing type.
The video above is not our final result as a group but a pick of my personal reflection on the programs used during the course and to show how the programs perform their functions.
The python script was presented by Xia Chen and Manav Mahan Singh (Ph.D. Candidates)
Machine learning in Architecture and Urban Design is a suitable method for prediction, performance tests, and optimization purposes.
The first method used in this experiment explored a decision tree approach. This method explored many phases such as problem definition, data exploratory analysis, feature engineering & model selection, hyperparameter optimization, result evaluation, and model explanation (SHAP) & visualization.
The second method explored the Deep learning method using the Keras library. The model was trained in the process and an optimized result was generated.
Note: The accuracy of the result generated in this experiment was not the best result because the dataset was very low.
However, it is evident that the higher the data, the higher the accuracy of the result.
#energySimulation #MachineLearning #ML #AI #MachineLearninginArchitecture #EnergyPerformance