In this paper, the Multi-Layer Perceptron (MLP) neural network is optimized with two metaheuristic algorithms, namely Artificial Bee Colony (ABC) and Particle Swarm Optimization (PSO) for estimating the heating load (HL) and cooling load (CL) of the energy efficient buildings with the residential use. To achieve this, a dataset composed of eight independent factors namely, relative compactness, surface area, wall area, roof area, overall height, orientation, glazing area, glazing area distribution, along with the two dependent variables of HL and CL is provided. Out of 768 samples, 80:20 ratio is considered to select the training and testing datasets randomly. Through a trial and error process, the optimal parameters of the MLP, ABC-MLP and PSO-MLP networks are determined. Three well-known criteria including coefficient of determination (R²), mean absolute error (MAE), and root mean square error (RMSE) are utilized to measure the accuracy of the used models. The results reveal that applying the ABC and PSO algorithms, helps the MLP to perform more efficiently. In this sense, the increase of R² from 0.8933 to 0.9349 and 0.9370 for the HL, and from 0.8872 to 0.8969 and 0.8997 for the CL prediction show that the outputs of the ensemble models (i.e., ABC-MLP and PSO-MLP) are more correlated with the actual data. Also, the MAE decreases as 22.32% and 24.28% for the HL, and 10.36% and 12.00% for the CL respectively by applying the ABC and PSO. Besides, the RSME decreases as 22.48% and 23.86% for the HL, and 6.06% and 7.56% for the CL modeling respectively with using the ABC and PSO. It is also deduced that the PSO outperforms the ABC in the performance enhancement of the MLP.

在本文中，多层感知器（MLP）神经网络使用两种元启发式算法进行了优化，分别是人工蜂群（ABC）和粒子群优化（PSO），以估算其的热负荷（HL）和冷却负荷（CL）具有住宅用途的节能建筑。为此，提供了一个数据集，该数据集由八个独立因素组成，即相对密实度，表面积，壁面积，屋顶面积，总高度，方向，玻璃面积，玻璃面积分布以及HL和CL的两个因变量。在768个样本中，考虑使用80:20的比率来随机选择训练和测试数据集。通过反复试验，确定了MLP，ABC-MLP和PSO-MLP网络的最佳参数。三个众所周知的标准，包括测定系数（R²），平均绝对误差（MAE）和均方根误差（RMSE）用于测量所用模型的准确性。结果表明，应用ABC和PSO算法有助于MLP高效执行。从这个意义上说，R ²的增加HL的从0.8933到0.9349和0.9370，CL预测的从0.8872到0.8969和0.8997表明集成模型（即ABC-MLP和PSO-MLP）的输出与实际数据更加相关。此外，通过应用ABC和PSO，HL的MAE分别降低了22.32％和24.28％，CL的MAE降低了10.36％和12.00％。此外，使用ABC和PSO时，HL的RSME分别降低了22.48％和23.86％，CL的RSME降低了6.06％和7.56％。还可以推断出，在MLP的性能增强方面，PSO优于ABC。