During the early design process, simulations allow numeric assessment and 3D models allow visual inspection for qualitative criteria. However, exploring different design alternatives based on both is challenging. To support the design exploration of quantitative performance and geometry typology of various design alternatives during the early design stages of indoor arenas, this paper proposed a novel design method of SOM-MLPNN by combing self-organizing map (SOM) and multi-layer perceptron neural network (MLPNN), based on the inspiration of local linear mapping based on self-organizing map (SOM-LLM). In SOM-LLM or SOM-MLPNN, the SOM can support designers to explore the whole design space according to geometry typologies and provides reference/labelled inputs for LLM/MLPNN to approximate multiple quantitative performance data for various design alternatives. Both SOM-LLM and SOM-MLPNN are applied and compared in a design of indoor arena. Besides the development of the method, original contributions include 1) proposing two operations (using a large size of SOM network and using a small amount of input data to train the SOM network) to save the computational time and increase the accuracy in data approximation and 2) proposing a series of data visualizations to interpret the results and support design explorations in different ways.
- Complex indoor arena
- Comprehensive design exploration
- SOM-LLM (local linear mapping based on self-organizing map)
- SOM-MLPNN (multi-layer perceptron neural network based on self-organizing map)