基于多传感器的室内人数检测模型

Model for Indoor Occupancy E stimati on Based on Multi-sensor

本文研究了数据挖掘在基于室内人数的建筑节能方向的应用,搭建了一套以无线传感器网络为基础的室内人数测量系统,采用基于相关性的特征选择算法从中选择与室内人数相关度最高的6类传感器组合,并通过支持向量回归、决策树、随机森林等数据挖掘方法进行了室内人数检测方法的研究,用以在不侵犯人员隐私的前提下实现室内人数的评估。结果显示三种方法均能有效推理室内人数,其中随机森林由于其集成特点使得基学习器之间得到互补,得到了最优的泛化性能,模型的均方误差、平均绝对误差、标准化平均绝对方差分别为0.007、0.006和0.083,优于其他两个模型。

Application of data mining in building energy conservation based on indoor population is studied in this paper.An indoor people counting system based on wireless sensor network(WSN)is built and used to collected the state features of the indoor environment.Correlation based Feature Selection is used to select the six sensor combinations with the highest correlation with the indoor population.Then the indoor people counting prediction method is conducted based on support vector regression,decision tree and random forest to provide support for decision-making of the air conditioning operation system.The results show that all the three methods are effective in inferring the number of people in the room,among which the random forest,due to its ensemble characteristics which makes the base learner complement each other,obtains the optimal generalization performance and the optimal inferring result.The MSE,MAE and NMSE of the model were 0.007,0.006 and 0.083,which were superior to the other two models.