Using Kriging regression to improve the stability and diversity in NSGA-II
International Building Performance Simulation Association ((IBPSA)
© 2017 IBPSA
Non-dominated sorting genetic algorithm version 2 (NSGA-II) is a multi-objective optimisation method. NSGA-II is often used to optimise the design of building. This paper details small improvements to this algorithm using ‘fitness approximation’ methods. Fitness approximation is used speed up the conversion of NSGA-II. Radial basis functions networks have been shown to be useful for this. Although there are many types of fitness approximation function that could be use for this purpose, Kriging methods have not yet been tested. In this paper, Kriging models are compared to standard NSGA-II. The results show that Kriging-based fitness approximation slightly improves upon standard NSGAII. More work is needed to test this method on different building types as well as more complex problems, such as those associated with HVAC design.
The authors would like to thank for EPSRC for funding this research [Ref: EP/M021890/1]
This is the author accepted manuscript. The final version is available from IBPSA
Building Simulation 2017, 15th International Conference of IBPSA, 7-9 August 2017, San Francisco, USA