Dr Sandy Brownlee – Loughborough University
Multi-objective optimisation of building designs
In recent years, evolutionary algorithms have increasingly been applied to the optimisation of real-world industrial problems. Optimisation of building designs is one such area: typical building designs have large numbers of variables, including construction materials, dimensions and equipment specifications. All of these can affect construction cost, operational energy use and occupant comfort.
The goal of Evolutionary Multi-objective Optimisation (EMO) is to find a set of designs representing a trade-off (or Pareto front) between conflicting objectives such as cost vs energy efficiency. This trade-off is used to support the end user in decision making. A typical building design optimisation problem will have two or three conflicting objectives and multiple constraints to be met. Further to this, a typical evolutionary algorithm evaluates thousands of solutions to converge on an optimum or optimal set; that building energy simulations can take minutes to hours represents a significant challenge. Techniques for overcoming this issue include multithreading, fitness inheritance and surrogate models of fitness.
In this presentation, I will introduce the broad concepts of EMO, and present recent work conducted as part of a project designed to deploy EMO techniques in a commercial software package. For two example building optimisation problems I will present some results, together with approaches we have taken in improving the algorithms to overcome challenges such as long computation time and constraint handling.
School of Computing, Robert Gordon University, St Andrew Street, Aberdeen, Lecture Room C48, 14:15 – 15:15.