【作者】 李力； 韩冬青； 董嘉；

【机构】 东南大学建筑学院；

【摘要】 建筑能耗不仅取决于建筑的物理节能性能,更受到使用者室内行为的影响。随着室内定位技术的发展,大量的室内行动轨迹可以被监测记录。这些轨迹既可以揭示人在室内的分布规律,也可以被进一步转译为具体使用行为,从而探究使用行为和建筑能耗间的关联性。但是,由于轨迹信息的复杂性、多样性及外部噪声干扰,现有统计分析方法很难对数据进行深度解读,且容易受到研究者先验知识的影响。本文提出了一种全新的基于卷积自编码神经网络(CAE)的行为轨迹聚类算法。该算法使用CAE模型,将原始轨迹数据进行压缩并产生对应的特征向量,根据特征向量间进行聚类分析。算法可以通过非监督式的学习方式,自动提取轨迹主要特征并排除噪音等干扰。此算法被成功应用于真实展览空间中参观人员行动特征的分析中,并成功提取了典型的空间行为轨迹模式。在此基础上,结合问卷调查的背景信息,将参观者人员的个体属性与行为模式进行了综合的分析,并发现内在规律。更多还原

【Abstract】 Building energy consumption not only depends on the physical energy efficiency of the building,but also by the user behavior.Whereas,due to the lack of efficient observation means,there is not enough data to support in-depth research.Now,With the development of indoor positioning technology, a large number of indoor action tracks can be recorded.These tracks can reveal the distribution of people in the room,or can be further translated into specific use behavior,In order to explore the use of behavior and building energy consumption correlation. More in-depth analysis of the study first need to be based on the similarity of action trajectory comparison.Whereas,because of the complexity and diversity of trajectory data,it greatly increases the difficulty and stability of data comparison.In this paper,a new algorithm based on convolutional self-coding neural network(CAE)is proposed The algorithm compresses the original trajectory data into corresponding feature vectors through the deep learning of CAE,The Euclidean distance between eigenvectors is used to compare the similarity of trajectories.The algorithm can be used in unsupervised learning,The main features are extracted automatically and the interference such as noise is ignored.This algorithm has been successfully applied to the analysis of the clustering features of visitors’,action trajectories in real exhibition space.Experimental results show that compared with the existing algorithms,the proposed algorithm has great advantages in robustness,flexibility and practicability.更多还原