A Data-Driven Modelling Approach for Large Scale Demand Profiling of Residential Buildings

In this paper the traditional use of data-driven models (DDM) as forecasting tools is coupled with parametric simulation to create a building modelling framework for demand profiling of a large number of buildings of the same typology. Most studies to date utilising DDM have been conducted on single buildings, with less evidence of the role that DDM may have as a modelling technique for application at scale. The proposed methodology is based on the use of a simulation-based building energy modelling tool and a parametric simulator to create a large dataset consisting of 4096 different building model scenarios. Three DDM techniques are utilised; Support Vector Machines, Neural Networks and Generalised Linear Models, these are trained and tested using the generated simulation dataset. Results, at an hourly resolution, show that DDM approaches can correctly emulate the outputs of the building simulation software with mean absolute error ranging from 4 to 9 percent for different DDM algorithms.