An efficient metamodel-based method to carry out multi-objective building performance optimizations

Abstract

Nowadays, performing multi-objective optimizations of actual building designs is one of the most challenging problems of the building energy efficiency area. This paper aims to propose an efficient method to solve multi-objective optimization building performance problems using a novel metamodel-based approach. To this end, the multi-objective Non-dominated Sorting Genetic Algorithm-II (NSGA-II) is dynamically coupled with artificial neural network (ANN) metamodels, which were previously trained with results of building performance simulations conducted using the EnergyPlusTM software. This new approach proposes an optimal way to generate the samples used to train and validate the ANN-based metamodels minimizing the total of building energy simulations necessary to train them, and guarantees accurate optimization results. To validate the strengths of the proposed method, it is applied to optimize the energy efficiency and thermal comfort of an actual dwelling in order to get the best trade-off (Pareto front) of the building between heating and cooling performance. This case study involves 12 of the more influential discrete and categorical design variables like roof types, external and internal wall types, solar orientation, solar absorptance, size and type of windows, and the dimension of external window shadings of this house among others, making a complex building performance optimization problem with more than 108 possibilities to choose. Furthermore, the results obtained are systematically compared and validated with the “true” Pareto front achieved using a simulate-based scheme which directly couples EnergyPlus program and NSGA-II algorithm. Results indicated that the presented method is able to reduce up to 75% the number of building energy simulations needed to find the Pareto front of an actual multi-objective building performance optimization problem, keeping a good accuracy of the results.