This paper introduces a novel building energy model reduction pipeline called ‘model-cluster-reduce’. It is centred around using clustering techniques to identify archetypes and eliminate redundant zones. An experiment was conducted in this paper using a detailed EnergyPlus model generated from building information modelling directly. A total number of four reduced models were generated and compared against the original model, random select models and an expert model. The reduced models estimated annual energy simulation and parametric simulation results within 5% error margin, while reducing the overall simulation time by 95%. The proposed method – which is aimed at large models where inter-zone heat transfer is not significant – can be used to approximate parametric simulations or optimizations with greatly reduced runtime.

Additional Metadata
Keywords building energy simulation, machine learning, model order reduction, optimization, parametric simulation
Persistent URL dx.doi.org/10.1080/19401493.2017.1410572
Journal Journal of Building Performance Simulation
Citation
Shi, Z. (Zixiao), & O'Brien, W. (2017). Building energy model reduction using model-cluster-reduce pipeline. Journal of Building Performance Simulation, 1–15. doi:10.1080/19401493.2017.1410572