Elsevier

Building and Environment

Volume 162, September 2019, 106280
Building and Environment

Cross-source sensing data fusion for building occupancy prediction with adaptive lasso feature filtering

https://doi.org/10.1016/j.buildenv.2019.106280Get rights and content

Highlights

  • Cross-source data fusion can improve occupancy detection accuracy.

  • Environmental and Wi-Fi prob sensing can be integrated with feature sets.

  • The adaptive lasso was implemented in feature selection.

  • CO2 concentration, temperature, and Wi-Fi signal indicators are the most correlated features.

Abstract

Fusing various sensing data sources can significantly improve the accuracy and reliability of building occupancy detection. Fusing environmental sensors and wireless network signals are seldom studied for its computational and technical complexity. This study aims to propose an integrated adaptive lasso model that is able to extract critical data features for environmental and Wi-Fi probe dual sensing sources. Through rapid feature extraction and process simplification, the proposed method aims to improve the computational efficiency of occupancy detecting models. To validate the proposed model, an onsite experiment was conducted to examine two occupancy data resolutions, (real-time and four-level occupancy resolutions). The results suggested that, among all twelve features, eight features are most relevant. The mean absolute error of the real-time occupancy can be reduced to 2.18 and F1_accuracy is about 84.36% for the four-level occupancy.

Introduction

HVAC (heating, ventilation, air-conditioning) systems consume over 40% energy in buildings and improving the energy efficiency of the HVAC systems is the key to minimize building energy consumption [1]. Occupancy is the core reference for the HVAC system operation, as HVAC systems are designed to maintain occupants’ thermal comfort and the indoor environment quality [2]. Also, since occupants can passively participate in building heating load transfer and actively control building facilities, occupancy information has significant impacts on building energy usage [[3], [4], [5]]. Therefore, occupancy detection and prediction have been recognized as one of the most inspiring topics for building environment and energy conservation [6,7].

Conventional occupancy detection and estimation rely on various environmental sensors, such as carbon dioxide (CO2) sensors [8], lighting sensors [9], passive infrared sensor [10], or wireless networks, such as Bluetooth [11,12], radio-frequency identification (RFID) [13], and Wi-Fi [14,15]. Many researchers proposed to integrate multiple sensors to improve the detection reliability [[16], [17], [18]]. For example, Masood et al. proposed a filter-wrapper component for a feature selection process to fuse the CO2 concentration, temperature, relative humidity, and air pressure level of commercial building spaces [19]. However, existing studies seldom investigate data fusion cross sensors and wireless network, as their heterogeneous technological basis and data formats. Therefore, how to select suitable data sources, determine proper data format, and develop feasible data fusion interface becomes the major technical barriers for cross-platform occupancy detection. This study took the environmental sensing and Wi-Fi network as the subjects and developed a quantitative method that can extract and fuse signal features for high-resolution building occupancy detection. The proposed method intends to optimize the feature selection and fusion with the physical balance model and an adaptive lasso method to improve the accuracy and reliability of the occupancy detection with predictive sensing outcomes.

Section snippets

Occupancy sensing with independent data sources

With the advances in electronics, various sensors and algorithms were developed and implemented in sensing building occupants. CO2 concentration is one of the most widely used parameters to guide the facility operation. For example, Wang et al. utilized physical balance function of dynamic CO2 concentration level to count a number of occupants in specific spaces [20,21]. Díaz and Jiménez applied CO2 to predict the occupancy pattern to match the computer power consumption [22]. Jiang et al.

Methodology

The proposed method can be divided into three major steps and Fig. 1 illustrates its schematic framework. The first step includes the infrastructure setup of the proposed method. The major infrastructure includes environmental sensing equipment, camera, Wi-Fi Probes, and database server. The second step implemented a feature abstraction process consisting of data preprocessing using an exponential moving average (EMA) filter. Physics-based models can be applied to extract features and select

Experiment setup

The test bed of the validation experiment is a graduate student office about 200 m2 with 25 residents. Fig. 2 shows the spatial layout of the test bed. The office has two entrances with overhead cameras installed. Both cameras recorded the entering and exit events of occupants and the records were used to generate ground truth with video analysis. Wi-Fi probes were installed to access the connection requests and responses of all wireless devices within the space. Several integrated environment

Results of the feature selection

Fig. 3, Fig. 4 present the feature selections for both real-time and four-level occupancy prediction datasets. With the adaptive lasso model, features that highly correlated to actual occupancy profiles were identified. Both occupancy resolutions have the same identified features. It also can be inferred from results that the outdoor air flow rate (Moa), and three related features filtered out. Additionally, the indoor air relative humidity (RHin), Moa, and outdoor air temperature (Tout) show a

Discussion

Building energy performance associate with the need and behavior of occupants, detecting and predicting accurate occupancy can significantly improve the operation efficiency of building facilities and promote their efficiency [54]. This study investigated the data fusion research for building occupancy prediction to figure out the better dataset combination and more suitable parameters through building operation and Wi-Fi datasets. Two kinds of occupancy information were selected in this study,

Conclusion

Data fusion technology with multiple sensors has attracted more and more attention in occupancy studies. This study proposed a data fusion study to integrate building physic-based, adaptive lasso, machine learning-based models for occupancy feature selection. This study defined two occupancy levels, real-time and four-level occupancy, and conducted one experiment to validate test occupancy feature selection process. In the results, a total of 12 features was selected from physics-based models

Acknowledgment

The work described in this study was sponsored by the projects of the National Natural Science Foundation of China (NSFC#51678127), the National Scientific and Technological Support during the 12th Five-Year Plan Period (No.: 2013BAJ10B13), and Beijing Advanced Innovation Center for Future Urban Design (UDC#016010100). Any opinions, findings, conclusions, or recommendations expressed in this study are those of the authors and do not necessarily reflect the views of the National Scientific and

References (57)

  • A. Szczurek et al.

    Occupancy determination based on time series of CO2 concentration, temperature and relative humidity

    Energy Build.

    (2017)
  • M.K. Masood et al.

    A novel feature selection famework with Hybrid Feature-Scaled Extreme Learning Machine (HFS-ELM) for indoor occupancy estimation

    Energy Build.

    (2018)
  • S. Wang

    Dynamic simulation of building VAV air-conditioning system and evaluation of EMCS on-line control strategies

    Build. Environ.

    (1999)
  • J.A. Díaz et al.

    Experimental assessment of room occupancy patterns in an office building. Comparison of different approaches based on CO 2 concentrations and computer power consumption

    Appl. Energy

    (2017)
  • J. Yang et al.

    Comparison of different occupancy counting methods for single system-single zone applications

    Energy Build.

    (2018)
  • Y. Benezeth et al.

    Towards a sensor for detecting human presence and characterizing activity

    Energy Build.

    (2011)
  • J. Page et al.

    A generalised stochastic model for the simulation of occupant presence

    Energy Build.

    (2008)
  • N. Li et al.

    Measuring and monitoring occupancy with an RFID based system for demand-driven HVAC operations

    Autom. ConStruct.

    (2012)
  • S. Salimi et al.

    Occupancy prediction model for open-plan o ffi ces using real-time location system and inhomogeneous Markov chain

    Build. Environ.

    (2019)
  • J. Chen et al.

    Assessing occupants' energy load variation through existing wireless network infrastructure in commercial and educational buildings

    Energy Build.

    (2014)
  • C. Martani et al.

    ENERNET: studying the dynamic relationship between building occupancy and energy consumption

    Energy Build.

    (2012)
  • J. Wang et al.

    Wi-Fi based occupancy detection in a complex indoor space under discontinuous wireless communication: a robust filtering based on event-triggered updating

    Build. Environ.

    (2019)
  • Y. Wang et al.

    Understanding occupancy pattern and improving building energy efficiency through Wi-Fi based indoor positioning

    Build. Environ.

    (2017)
  • W. Wang et al.

    Modeling and predicting occupancy profile in office space with a Wi-Fi probe-based Dynamic Markov Time-Window Inference approach

    Build. Environ.

    (2017)
  • L.M. Candanedo et al.

    Accurate occupancy detection of an office room from light, temperature, humidity and CO2 measurements using statistical learning models

    Energy Build.

    (2016)
  • B. Dong et al.

    An information technology enabled sustainability test-bed (ITEST) for occupancy detection through an environmental sensing network

    Energy Build.

    (2010)
  • T.H. Pedersen et al.

    Method for room occupancy detection based on trajectory of indoor climate sensor data

    Build. Environ.

    (2017)
  • J.P. Roselyn et al.

    Development and implementation of novel sensor fusion algorithm for occupancy detection and automation in energy efficient buildings

    Sustain. Cities Soc.

    (2019)
  • Cited by (27)

    • A real-time occupancy detection system for unoccupied, normally and abnormally occupied situation discrimination via sensor array and cloud platform in indoor environment

      2021, Sensors and Actuators A: Physical
      Citation Excerpt :

      Computer vision technology based-methods have to consider high computational complexity, privacy issues, low image quality of surveillance etc. Integrated multiple sensors have been proposed to improve the detection reliability [16]. Sensor array fusion is a technique that combines data collected from multiple sensors to provide enhanced performance over any single sensor [17–19].

    View all citing articles on Scopus
    View full text