Building energy optimisation using artificial neural network and ant colony optimisation

Reducing energy consumption is one of the world’s most challenging issues particularly with increases in population and economic growth. Building optimisation methods have proven to be powerful tools for minimising building energy consumption. However, conventional optimisation methods (e.g. parametric studies or simulation-based-optimisation methods) are extremely time-consuming due to inherent properties of stochastic optimisation algorithms and highly-nonlinear thermal behaviour of buildings. To address this issue, an artificial neural network method is developed in this research to create a meta-model (surrogate model) of a typical commercial building in Australia. This meta-model is then optimised with respect to different building design variables (e.g. windows shadings size) using Hybrid Ant Colony Optimisation and a gradient based algorithm (Interior Point Algorithm) to ensure the optimality properties of solutions. A benchmarking study is conducted comparing existing software-in-the-loop optimisation methods (Particle Swarm and Ant Colony Optimisation) with the surrogate model-based approach. The results demonstrate that building optimisation using meta-models can find the optimal building design with less computational burden, and identify the potential energy saving solutions in different climates in Australia. Additionally, analysis of the optimisation results provides a better understanding of optimal values of design variables in Australian climates, and help building designers set up future building codes to design high performance buildings.