Paper Conference

Proceedings of Building Simulation 2017: 15th Conference of IBPSA

     

Development of Energy Simulation Models from Smart Meter Data using Inverse Modelling and Genetic Algorithms

Daniel Costola1, Ana Paula Melo2 and Loïc Jacob1
1Department of Mechanical and Aerospace Engineering, University of Strathclyde, Glasgow, G11XJ, UK
2Laboratory for Energy Efficiency in Buildings, Federal University of Santa Catarina, Brazil


DOI: https://doi.org/10.26868/25222708.2017.701
Abstract: To combat anthropogenic climate change the energy use of buildings must be reduced significantly within the coming decades. There is a pressing need for cheap and accurate baseline building energy models to inform owners on opportunities for energy efficiency improvements. The increasing penetration of smart meters in existing buildings provides a wealth of data that can be leveraged with dynamic simulation models to achieve energy savings. This paper describes a method to generate physically driven dynamic simulation models from metered data using inverse modelling techniques. The application of inverse modelling methods to generate baseline building energy models using a physically driven software is novel in the field, with potential for cheap prediction of the impacts of energy efficiency upgrades in stock modelling. The tools used are the dynamic simulation software ESP-r and the genetic algorithm from MATLAB’s Global Optimisation Toolbox. Building data and gas meter readings for the year of 2013 provided by the University of Strathclyde were used to drive the modelling of an office building in the campus. Results show that the optimised models have an energy consumption with 5% difference to the metered data provided. Convergence can be achieved within a reasonable number of generations and with a population size that is not prohibitively large. However, adding variables significantly increases the computation time required for convergence. Further work could explore the limitations of the method when applied to complex models with higher number of variables.
Pages: 2475 - 2483
Paper:
BS2017_701