Neural Networks as Decision Support System for Energy Efficient Building Design

This paper outlines the possibility of the application of neural networks to energy efficient building design. The case of thermal design of building elements has been chosen as an illustration. The neural network employed for this purpose is feed forward, non-recurrent, and multilayered. The training of neural network is based on the back propagation algorithm. The network was trained for an input set of desirable thermal parameters namely thermal transmittance, time lag and decrement factor. The output parameters constituted the material and section properties of the building element. Once trained, the net was then tested for the patterns for which it had not been trained, and the error percentage was found to be within a permissible limit.