The complexityof MOEA is that there is a competitive relationship between eachobjective function, that is, when one of the objective functionsachieves better results the others optimization results may not be realistic. Therefore, the optimal solution is often not the only one,but a set of optimal solutions which do not have mutually domi-nated relationship, that is, the non-dominated front. The algorithmemploys the Pareto (front) which is used to calculate the fitness ofeach solution (solutions of equal rank having equal fitness).This paper adopted the non-dominated-and-crowding sortinggenetic algorithm II (NSGA-II), developed by Deb [18]. This algo-rithm uses a specific population sorting approach, which is firstlybased on dominance, and then on a crowding distance computedfor each inpidual. Due to this selection process, both convergenceand spreading of the solution front are ensured, without requiringthe use of external population. NSGA-II is capable of maintainingthe population persity and avoid the loss of excellent inpidualsat small computation, and there is no need to set some parame-ters of algorithm (such as the share and mate in MOGA). NSGA-IIis recognized as one of the most efficient MOEA.2.2. Objective of fitness function–back propagation neuralnetwork combined with genetic algorithmThere are many factors that affect building energy consump-tion. These include transparency ratio (%), building form factor,orientation, optical and thermo-physical properties of the materialsused in building envelope etc. These factors have nonlinear cou-pling impact on building energy consumption. Therefore, dynamiccomputing is the main energy analysis method in the process ofbuilding efficiency design. Due to the complexity of the dynamiccomputing process, it is difficult for the general engineer staff tomaster it. In general, most of the commercial dynamic modelingprograms are time-consuming, especially when it comes to provid-ing the results annually. Furthermore, the cost of these programs isprohibitive for small research establishments. Therefore, there is aneed for alternative approaches to perform this task. As a result, thealternative approaches must have high accuracy and computationalspeed which can greatly simplify the design optimization processand reduce design optimization time in building energy efficiency.The recently developed technology, artificial neural network (ANN)could offer such an alternative approach [19].Therefore, ANN is used to establish the multi-objective pre-diction model and the Genetic algorithm for optimizing the ANNis used to improve the prediction accuracy. Dorsey et al., [20]designed a genetic adaptive neural network training (GANNT) algo-rithm and shown that the GA also worked well for optimizing theANN. The GANNT algorithm is different from other genetic searchalgorithms because it uses real values instead of binary representa-tions of the weights. In this paper, the genetic algorithm is adoptedfor training the BP network. The genetic algorithm is utilized tooptimize the BP network’s weight or threshold. Fig. 1 illustrates asimple outline of the GA used in this paper [21].3. The multi-objective optimization model in buildingsustainable design3.1. The multi-objective optimization model3.1.1. Components of the optimization modelThe components of the optimization model are presented in thefollowing order: variables, constraints, and objective functions. Themodel concentrates on building conceptual design because of itsimportance in determining the performance of both energy con-sumption and of indoor thermal comfort. The same methodologycould be applied later to a large scope covering other building sys-tems such as heating, ventilation, and air conditioning system.• Variables: In this study, the buildings are limited to a rectangularshape with a known total floor area. 办公室空调设计英文文献和中文翻译(4):http://www.youerw.com/fanyi/lunwen_29101.html