A neural network solution to an architectual design problem: Design of a light shelf

The analysis and evaluation of light-shelves design are complex and involve numerous parameters. These features require the development of various methods and tools to help the designer analyse and evaluate the many possible configurations. Conventional statistical tools for generating performance algorithms are not easily adaptable to this problem due to the very large number of independent design variables involved in the equation. In this article Neural Networks (NNs) are introduced as a promising computation method for the analysis and evaluation of daylight illumination due to the light shelf. NNs are shown to have powerful capabilities for generalising a solution by learning a few examples, by associative and adaptive processing, and tolerating the fault inherited from incomplete data.