论文推送| 基于模糊C均值聚类算法的传感器部署策略:面向室内环境安全与健康的在线控制
基于模糊C均值聚类算法的传感器部署策略:面向室内环境安全与健康的在线控制
Sensor deployment strategy using cluster analysis of FuzzyC-Means Algorithm: Towards online control of indoor environment’s safety and health
Authors: Shi-Jie Cao, Junwei Ding, Chen Ren
Sustainable cities and society
https://doi.org/10.1016/j.scs.2020.102190
摘 要
Sensor monitoring plays a very important role in indoor environment control and energy efficiency. However, it is always a challenging task when discussing the methods and strategy for sensor deployment, considering its numbers and locations. Thus, this work aims to provide a systematic methodology of sensor deployment for efficient indoor environment monitoring by using clustering method of Fuzzy C-means (FCM) algorithm. An experimental chamber was considered for ventilation basis. Low-dimensional Linear Ventilation Models (LLVM) were applied to generate a series of low-dimensional pollutant concentrations, representing the potential hypothetic monitoring data. FCM algorithm was then adopted to cluster these data as well as corresponding grid points classified by three zones, i.e., main flow, main diffusion and well-mixed zones. Then the cluster centers of each zone were determined representing as the locations of sensor deployment. Lastly, taking the cluster centers as the hypothetic sensor locations and corresponding concentrations as the input for LLVM-based ANN (Artificial Neural Network), the predicted volume-averaged CO2 concentrations agree well with the simulated ones directly from low-dimensional CFD results, with the maximal error of 6.5 %. This work will be further facilitating the realization of artificial intelligent ventilation system and efficient indoor environment control.
研究背景
室内空气品质(IAQ)影响公众的生活质量和健康。暴露于空气污染物(高浓度的二氧化碳、挥发性有机化合物、颗粒物或微生物污染物)可能导致急性的过敏反应或慢性的健康问题。一些突发性的公共卫生事件(甲类传染病、生化武器等),污染物在大空间内快速扩散,如果不能有效控制将严重危及室内居民的健康安全。传统的建筑通风系统往往使用单一模式的气流组织并且在设定的通风速率下运行,消耗大量能量的同时还不能保证污染物的有效去除,不利于构建安全、健康、节能的建筑环境。随着物联网监测、低成本传感器技术等新兴技术的快速发展,传感器监测在室内环境参数(温度、湿度、污染物浓度)控制和建筑节能方面起着越来越重要的作用。任宸、曹世杰提出由 “监测”、 “预测”和 “控制”三个模块组成的智能通风系统,利用有限传感器获取的监测数据作为快速预测模型输入,实现室内环境参数的快速预测和HVAC系统的在线控制。
但是,一个可靠、高效的室内传感器网络(服务于智能通风系统)仍然面临一些挑战。关于建筑空间中传感器的部署策略,传感器数量和位置的设置一直是一个比较难确定的问题。一是在室内环境中需要部署多少个传感器。虽然室内污染物通常随时间和空间而变化,但考虑到成本和实际限制,通风控制系统通常在混合良好的区域(如靠近出风口)使用单个传感器进行评价。当涉及到一个复杂的系统时,往往需要部署更多的传感器来获取更精确的室内环境的时空数据,但是过多的传感器又会增加初期的投入成本和数据处理的负荷。此外,传感器的位置也需要重点考虑。任杰,曹世杰提出了传感器部署的经验策略,即传感器应避免沿同一或平行于主流平面布置;传感器应放置在“充分混合区”(接近出风口的位置)或附近,且应远离送风口。
因此,本工作旨在以有限监测点重建室内污染浓度分布为导向,以实现实时控制为最终目标,为室内环境监测提出了一种基于模糊聚类算法的传感器部署系统方法。该工作将进一步促进人工智能通风系统的实现和室内环境的高效控制。
研究内容
通过“监测”、 “预测”和 “控制”这三个步骤,在快速预测和评估不同通风方式下污染物非均匀的浓度分布的基础上,进行通风方式的快速决定性调整。因此,我们利用有限的在线监测数据,将在线监测浓度作为人工神经网络训练和预测的输入,进一步发展上述模型,实现对室内污染物分布的快于实时预测。
通过使用低维线性通风模型(LLVM)生成不同工况下的低维污染分布浓度进行模糊聚类分析(FCM),将室内的送风环境分为送风主流区域、污染扩散区域和充分混合区域三个区域。我们的快速预测模型将污染源信息(位置和强度)作为ANN训练的参数,以聚类中心作为传感器位置,相应的浓度数据作为快速预测模型(LLVM-based ANN)的输入,预测室内二氧化碳浓度的低维分布并与实际的CFD模拟数据进行了误差分析。
Fig. 1 Schematicrepresentation of LLVM-based FCM (dots circle Red represents the clustercenter).
Fig. 2 Theclustering results of hypothetic monitoring points and two categorized caseswith the pollutant source position of A and ABDE, divided by (a) 3 × 3 × 3
cubes, Row 1 and3; (b) 5 × 4 × 3 cubes, Row 2 and 4. (Cluster center is lined by red and zone labelindicated above).
研究结果
结果显示将模糊聚类结果作为快速预测模型的输入,污染物浓度非均匀非均匀低维分布与实际的CFD模拟结果最大误差为6.5%。
Fig. 3 Comparisonof low-dimensional CO2 concentration fields obtained from LLVM-based ANN (usingsensor data in the monitoring zones as inputs) and CFD
validation whenACH equals to 11 [(a): the case with source position A, 3 × 3 × 3; (b): thecase with source position ABDE, 3 × 3 × 3; (c): the case with source
position A, 5 × 4× 3; (d): the case with source position ABDE, 5 × 4 × 3].
文章信息
Shi-Jie Cao, Junwei Ding, Chen Ren. Sensor deployment strategy usingcluster analysis of Fuzzy C-Means Algorithm: Towards online control of indoorenvironment’s safety and health. Sustainable Cities and Society, Volume59, 2020, 102190, https://doi.org/10.1016/j.scs.2020.102190.