人工神经网络在绿色建筑评估中的应用

Artificial neural network in green building assessment

除了线性关系,绿色建筑评估的各级指标与整体性能之间还存在着非线性关系,必须正确反映这些关系以帮助相关组织和个人对绿色建筑的进行甄别。参照《绿色奥运建筑评估体系》设计阶段的指标体系,使用人工神经网络模拟评估人的评估决策,采用评估人给出的40个建筑物评估数据用于训练和仿真。根据指标的逻辑关系采用"级联"方式构建神经网络,分两个阶段进行训练和仿真,降低了小样本对训练效果的不利影响。对那些均方误差已经满足要求的网络,利用确证样本验证其泛化能力。实例显示构建的人工神经网络的学习、泛化能力较强,误判率较低。

Besides linear relationship, non-linear relationship is found between the indicators of green building assessment and its function. The paper has tried to correctly figure out these relationships so as to help those invovled to distinguish better buildings. Based on The Assessment System for Green Building of Beijing Olympics, the authors have fabricated an artificial neural network to simulate experts＂ decision on assessment by assessing 40 buildings to provide samples for training, verifying and simulating. According to the logical structure of indicators, a ＂cascade＂ model is introduced in building neural networks which are trained and simulated in two stages, reducing the adverse impact produced by the small sample in training. Verifying samples are used to test generalization of the networks whose MSE achieved the given goal. The simulation results show that the proposed model has a strong ability to learn and generalize information implied in the training sample with less misjudgment.