Paper Conference

Proceedings of eSim 2022: 12th Conference of IBPSA-Canada

     

Using a Convolutional Neural Network to Determine the Thermal Characteristics of a Building

Liam Jowett-Lockwood, Ralph Evins
University of Victoria, Canada

Abstract: Surrogate models are machine learning models that are trained using detailed simulation input and output data and can practically provide instant results. They use a forward implementation strategy and normally do not use building simulation outputs to identify inputs. This paper introduces inverse surrogate modeling of a building by using a convolutional neural network that uses temperature data to estimate the building characteristics, such as wall insulation conductivity and infiltration flow rate. The training data resembles thousands of 10-minute interval temperature time series along with their respective building parameters. The first proof of principle uses synthetic data from a building energy model to train and test the neural network. This will later predict actual building parameters using real temperature data from an existing building. Findings demonstrate that combining temperature data with a convolutional neural network could serve as a cost-effective method of building parameter identification.
Keywords: Surrogate-modeling, Simulation, Temperature, Convolutional, Inverse
Paper:
esim2022_259