SEED: Public Energy and Environment Dataset for Optimizing HVAC Operation in Subway Stations

Technical Summary: SEED: Public Energy and Environment Dataset for Optimizing HVAC Operation in Subway Stations

Problem Statement
Optimizing Heating, Ventilation, and Air Conditioning (HVAC) systems is critical for energy sustainability, as these systems consume a significant portion of total power in commercial buildings and subway stations (often exceeding 40%). Effective optimization requires three steps: identifying thermal and HVAC signatures, designing optimal control policies, and evaluating those policies. Currently, this process is hindered by a lack of publicly available, fine-grained datasets. Existing approaches rely heavily on theoretical models or simulations (e.g., DeST, EnergyPlus), which are time-consuming to configure and often lack fidelity due to the diversity of building structures and the difficulty of monitoring internal HVAC states. While datasets exist for smart home energy disaggregation (e.g., REDD, BLUED), there is a scarcity of real-world, long-term data regarding the working states, thermal environments, and user dynamics specific to subway HVAC systems.

Methodology
The authors present the Subway station Energy and Environment Dataset (SEED), collected from multiple stations on a Beijing subway line during August and September 2013. The data collection infrastructure was developed as part of a project to create autonomous HVAC control systems for the Beijing metro.

• Sensor Deployment: The system monitors a transfer station (anonymized for privacy) using a distributed sensor network.
  • Environment: Temperature, humidity, and CO2 sensors are deployed at four indoor and two outdoor locations, reporting data at one-minute resolution.
  • Passenger Flow: Hourly passenger flow data (check-in and check-out counts) is acquired from the subway's ticket checking system.
  • HVAC Systems: Power meters and internal sensors monitor the real-time states of refrigerators, supply/return/exhaust fans, cooling/chilling pumps, and valves. Data is reported at one-minute resolution.
• System Architecture: The HVAC systems utilize Carrier equipment connected via a Profibus DP network to an information collection server. A custom control cabinet and autonomous logic were developed to manage the system.
• Data Scope: The dataset includes minute-resolution environmental and HVAC operational data, alongside hourly passenger flow data, covering periods of varying outdoor temperatures (from very hot days to cooler days).

Key Contributions

1. Public Dataset Release: The primary contribution is the release of SEED, a comprehensive dataset containing environment dynamics, HVAC working states, energy consumptions, and passenger flows. The dataset is available online to facilitate research in HVAC optimization.
2. System Description: The paper details the hardware and software architecture used for data collection and environment control in a real-world subway setting, providing a reference for similar deployments.
3. Initial Investigations: The authors utilize the dataset to perform preliminary analyses on:
   • Energy Disaggregation: Separating HVAC energy consumption into stable ventilating loads (fans) and dynamic cooling loads (refrigerators and pumps).
   • Thermal Signatures: Analyzing the relationship between outdoor/indoor temperature differences and refrigerator operating states.
   • Passenger Flow Signatures: Characterizing flow patterns based on time (workdays vs. weekends) and location (CBD vs. town center).
   • CO2 Correlation: Examining the correlation between CO2 density and passenger flow.
   • Response Latency: Evaluating the speed at which indoor temperature responds to cooling supply changes.

Results and Observations

• Energy Consumption Patterns: HVAC energy consumption is highly correlated with outdoor temperatures. While fan consumption remains relatively stable across different weather conditions, the energy usage of refrigerators and pumps varies significantly, suggesting that optimizing cooling utilities is the primary avenue for energy savings.
• Refrigerator Dynamics: The working loads of refrigerators closely respond to the temperature difference between indoor and outdoor environments.
• Passenger Flow Signatures: Passenger flow patterns differ significantly between workdays and weekends. Workdays show sharp peaks during rush hours, particularly at stations near business districts (CBD), whereas weekends show smoother curves with peaks in the afternoon (15:00–16:00), especially near leisure centers.
• CO2 and Flow Correlation: CO2 density variations closely track passenger flow variations. This suggests that CO2 data could potentially be used to infer passenger numbers if direct flow data is unavailable.
• Thermal Response: The indoor temperature responds quickly to changes in the cooling air supply temperature, indicating that the control latency in these specific subway HVAC settings is relatively short.

Significance
The paper claims that the availability of SEED will benefit the research community by enabling the identification of thermal environment and HVAC signatures without the need for expensive and fidelity-limited simulations. By providing real-world, fine-grained data, the dataset aims to support the design and evaluation of optimal control policies for subway HVAC systems, ultimately contributing to energy conservation and sustainability in public transportation infrastructure. The authors position the dataset as a foundational resource for future work in load signature identification and control policy optimization.