Sage Journals: Discover world-class research

Abstract

Arrivals and departures lie at the intersection of travel and building occupancy behaviours which dominate the landscape of energy demand in urban areas. Although transport and building systems are clearly linked, existing studies rarely consider the interactions between these systems in their modelling frameworks, thus restricting the policy-relevant scenarios that can be tested. This paper contributes to the field of data-driven energy modelling by proposing a flexible framework to integrate the modelling of travel and building occupancy behaviours, in which a travel simulator is coupled with a building occupancy model through a proposed mesoscopic link. The framework is operationalised in the context of the South Kensington Campus, Imperial College London, using the UK Time Use Survey data and Wi-Fi traceable logs. Implementing the framework for a hypothetical transport incident (i.e. sudden closure of the nearest underground station) generates people’s occupancy and circulation patterns across buildings, thus providing actionable insights for district-level smart grid planning and management. From a district planning perspective, occupancy schedules and dynamics in closed buildings are sensitive to incidents, whereas open and shared buildings are relatively stable. This finding indicates the need for flexible energy controls and smart grids with energy storage. From a building management perspective, occupancy durations generally reduce when affected by incidents, suggesting shortening the schedules of heating, ventilation and air-conditioning systems. From a facility management perspective, big changes in occupancy of closed buildings indicate unstable demands for the surrounding equipment (e.g. e-scooters, chargers), and efficiencies may be gained by allocating spaces/schedules to meet the dynamic demand.

Keywords

Timing of travel building occupancy local transitions smart grids energy management

Introduction

Transport and building systems are dominant drivers of energy use in urban areas, greatly impacting the urban environment and leading to climate change (Hickman and Banister, 2014). The building sector is a substantial energy consumer, responsible for nearly 35% of global energy use, while nearly 15% of CO₂ emissions are produced due to its operation and construction (IEA, 2021). The transport sector consumes around 28% of global energy (IEA, 2021) and produces larger global shares of CO₂ emissions at about 25%, which keeps growing at an average annual rate of 1.7% (IEA, 2022). In response, decentralised and smart solutions have been proposed to distribute energy controls in urban systems; enabling the integration of intelligent energy devices and intermittent renewable energy sources to reduce CO₂ emissions and deal with the uncertainties related to climate change (Wieczorek et al., 2024). Furthermore, the paradigm of data-driven energy management emphasises its role in providing decision support for all stakeholders in smart grids, including producers, operators, customers and regulators (Zhou et al., 2016).

Energy demand models for transport and buildings have been conventionally handled separately (i.e. building estates vs transport departments). With widespread electrification of transport, smart grids have begun to emerge and there are attempts to integrate power infrastructures of transport and building systems at the district level. However, there remains a gap in combining the demand models for the two systems, with insufficient understanding of their planning and operation strategies within integrated smart grids. From the perspective of the consumers (i.e. users of building and transport systems), the majority of their daily time is spent engaging in activities within buildings, whilst travel serves as the link connecting activities that occur in various buildings spread across the urban area (Rakha et al., 2014). These spatial and temporal connections indicate the presence of important interactions between transport and building systems from a modelling perspective. Specifically, traffic conditions may be associated with building occupancy as high building occupancy levels could indicate high demand for outbound and/or inbound travel from/to the building; traffic congestion could, in turn, impact arrivals at buildings, especially when arising unexpectedly (preventing the travellers from accounting for it in their travel plans). In order to combine the demand models for transport and building systems, we need a systems approach that can account for the interactions between activities in transport and building systems. Such an integrated framework can describe energy demand and flows within the urban area more effectively, further promoting the planning and operation of smart grid strategies.

Recently, a few studies have attempted to integrate the modelling of activities in transport and building systems in order to generate energy demand simulations (Rezvany et al., 2021, 2023). Earlier studies in this research area (Happle et al., 2020; Rakha et al., 2014; Wang et al., 2020) have explored the use of travel data to generate occupancy patterns for multi-building modelling, creating a landscape for urban planning with the energy perspective. However, the transport and building systems are typically treated as exogenous inputs to each other, capturing the effects of one system on the other (Hou et al., 2022). Such approaches are often motivated by a single-faceted focus on either buildings (occupancy, energy consumption) or transport (demand, operations). Overall, the temporal correlations between travel and building occupancy activities have been overlooked, potentially leading to the biased simulation of energy demand. For example, while building operation strategies commonly follow regular occupancy schedules, late arrivals at one building due to large-scale travel disruption could cause an unnecessary oversupply of building energy. Taking COVID-19 as another example, people were encouraged to stay at home and reduce their travels, which would have led to shifts in energy use from travel to building uses as well as spatial changes in building energy use patterns. It is therefore important to investigate how the two modelling domains (transport and buildings) can be integrated to capture the interactions between their activities and demands, to ultimately support better decision-making in planning and operations.

Towards that end, this paper aims to present a framework that shows how transport and building systems models can operate in an integrated manner within the same urban context, enabling the analysis of their mutual effects. The framework consists of three models at macroscopic, mesoscopic and microscopic levels, though this paper puts most emphasis on the linking element at the mesoscopic level, with the joint simulations using examples of established macroscopic and microscopic models (see previous studies of Hou et al., 2022; Pawlak et al., 2021, 2022). Notably, the empirical analysis is presented to demonstrate the framework application from transport to buildings at the district level and across one full day. As discussed in the literature review, to the best of the authors’ knowledge, this is the first attempt at developing a joint travel and building occupancy demand model, which is also implemented in a real-world empirical context. Implementing the model in specific scenarios of underground metro station closure shows how transport conditions can affect building systems, thereby providing evidence-based information to support the decision-making in future district-level smart grids.

This paper is structured as follows. First, we briefly review the literature on building occupancy modelling and the use of Wi-Fi data, especially in relation to the activity-based modelling (ABM) paradigm. Next, we introduce the conceptual and operational modelling frameworks. We then describe the empirical case study context, data and model implementation details. Following this, we present the empirical results, while discussing the potential applications of the proposed framework and its current limitations. Finally, we summarise the contributions of this study and highlight its potential impact on future research.

Existing studies

Building occupancy simulation for higher operational efficiency

Building occupancy modelling uses mathematical models to assist in interpreting and predicting people’s stay and circulation patterns within building spaces. In the building context, an occupancy pattern (or schedule) not only reflects the times of first presence (arrival at the building) and last absence (departure from the building) but also the variations in the number of occupants over time at an indoor zone of the building. Incorporating occupancy patterns into building design and management can lead to improved building energy efficiency (Yan et al., 2017), as it helps eliminate unnecessary energy consumption during unoccupied hours. With this perspective, accurate modelling and a comprehensive understanding of building occupancy can provide better insights into energy demand for plug loads as well as heating, ventilation and air conditioning (HVAC) systems in a building. Furthermore, understanding indoor activities, which impact building occupancy in response to incentives (e.g. time-varying energy tariffs), can help design policies that shift the energy demand from peak hours (Pawlak et al., 2021), thereby ensuring reliable energy supply.

To achieve higher energy efficiency, building occupancy simulations are integrated with the development of building codes and standards as well as facility control such as HVAC systems. The traditional strategies of building design and control assume building occupancy patterns to be the static full occupancy specific to building functionality (e.g. office, residence, etc.). In reality, occupancy within a building and in a group of buildings interacts with the surrounding environment, including the transport system. For example, traffic congestion or delays on public transport networks could lead to late arrivals at trip-destination buildings and subsequent changes in indoor occupancy; in turn, occupancy affected by external transport conditions will impact departures from the occupied buildings for the subsequent travel episode. Using the example of a departure from a large event (Giuliano and Lu, 2021; Kwoczek et al., 2014), an extension of the event (such as extra time in a game) can shift the surrounding traffic congestion to a later time. Looking at the reverse impacts, substantial congestion on the transport network may cause some visitors to arrive late at the event’s venue, changing the occupancy patterns at the venue.

Generally, three categories of models have been proposed in the literature for building occupancy simulation: stochastic, econometric and machine learning models (Hou et al., 2022). Most approaches are derived from the stochastic discrete-time Markov Chain model (Chen et al., 2015; Page et al., 2008) due to its effective performance in capturing occupancy variations over time. Hou et al. (2022), on the other hand, adopted an econometric approach and proposed a nested hazard-based model (NHBM), for which urban context was integrated into building occupancy simulations, enabling the incorporation and interpretation of transport impact on building occupancy (though not vice versa). Their model contributes to understanding building occupancy and people’s circulation across buildings on campus, thus providing evidence-based information for improving building energy efficiency. Machine learning algorithms have presented their advantages in short-term forecasting (e.g. decision tree by D’Oca and Hong, 2015) to achieve accurate occupancy predictions. In the present study, we rely on the NHBM approach (Hou et al., 2022) as it provides a feasible and flexible approach to incorporating transport-related variables into occupancy simulations.

Activity-based modelling in relation to building occupancy simulation

The idea of linking travel-oriented activity-based models (ABMs) to occupancy simulation is relatively recent. Mosteiro-Romero et al. (2020) adapted the MATSim (Axhausen et al., 2016) ABM suite to scale up the simulation of building occupancy schedules to a district level. Theirs was one of the first attempts to adapt transport models to address building simulations. However, their simulator was only used to replicate occupancy schedules randomly for building energy analysis, rather than modelling interactions/interdependence between the two systems (and enabling evaluations of the policy impact of one system on the other). Rezvany et al. (2021, 2023) took into consideration the interdependence of travel behaviour and in-home and out-of-home building energy demand in generating the activity-travel plan for an individual. Their framework includes a connection between activities in domestic buildings and outdoor travel but lacks explicit descriptions of how behaviour in non-domestic buildings (e.g. office buildings) links with travel. In another practical study, Pawlak et al. (2021) adapted an ABM of travel demand to achieve the detailed modelling and simulation of household activities in domestic buildings and their associated energy demand.

Among the attributes of an activity-travel plan, trip departure times are critical factors that can drive conditions on the transport network, in particular, challenges to the capacity of transport supply described in terms of the ability to accommodate demand. As the interest is in when the demand occurs, modelling the timing of travel (or time of departure) has been a research stream in the transport domain since at least the 1990s. For example, existing studies have demonstrated that peak departure from a large event (e.g. football match or concert) with a fixed end-time can cause surrounding traffic congestion (Giuliano and Lu, 2021; Kwoczek et al., 2014). Therefore, the proper use of departure distributions from an event venue would help to improve urban traffic management, reducing delays and avoiding accidents during severe congestion. Similarly, a business area that accommodates many people in office buildings with flexible but similar departure times is also highly likely to lead to traffic congestion (De Palma and Rochat, 1999).

Given this context, monitoring arrivals and departures in a group of buildings presents a novel opportunity to estimate real-time travel demand, thereby not only assisting with traffic management but also contributing towards the design of mobility services, such as taxi sharing or on-demand bus services. However, a comprehensive understanding of the relationship between building occupancy behaviour (i.e. arrival, departure, and stay) and outdoor travel behaviour is still lacking, for example, interactions between transport and building systems on future decentralised networks. This argument underpins a research need for the integrated analysis of building occupancy and activity-travel behaviour. To address the research significance, this paper proposes a novel framework for linking indoor and outdoor activities at an individual level, of which the indoor movement process (e.g. taking an elevator) is ignored due to its small share on an individual’s activity chain.

The use of Wi-Fi data for occupancy and mobility modelling

Since the Internet of Things (IoT) and wireless Internet have become available, new opportunities for sensing and measuring urban activities have emerged in the literature (Salim et al., 2020). Wi-Fi plays a vital role in delivering IoT innovation and has been integrated into mobile devices with network facilities broadly installed in indoor spaces. Given the advantages of indoor positioning, Wi-Fi logs have been deployed in past studies to monitor people’s occupancy and mobility, using smartphones as a proxy for user location. For example, Kalogianni et al. (2015) and Bon et al. (2016) employed traceable Wi-Fi data to identify movement patterns between buildings within a campus, whereas Griffioen et al. (2017) further specified the movement flows in corridors between buildings for evaluating the facility uses. The findings of these studies support the estate management team in the decision-making for facility planning and renovation. As COVID-19 restricted people’s travels and encouraged them to stay within closed spaces, Wi-Fi data presents its advantages of tracing people’s indoor occupancy and mobility. Zakaria et al. (2022) adopted the Wi-Fi data to characterise occupancy and mobility patterns under different COVID-19 policies (e.g. online learning and split-team) on campuses and analyse the policy impacts. Beyond the campus scenario, Traunmueller et al. (2018) used a large-scale Wi-Fi probe to model urban mobility trajectories in dense urban environments and identify usage intensity levels for street segments. These studies have demonstrated that Wi-Fi data is a feasible source for measuring people’s occupancy and mobility at the district level. Accordingly, this paper will employ Wi-Fi data as the dominant input for the implementation of the proposed district-scale model.

Systems approach for integrated modelling

A conceptual framework for integrating models of travel and building occupancy

We propose a conceptual framework shown in Figure 1 to elaborate on the links between the modelling of travel patterns and building occupancy behaviour. In an urban context, buildings provide indoor spaces that motivate mobility and activities on a smaller scale, whereas transport systems focus on outdoor mobility on a larger scale. Of course, activities can also take place outdoors, but this can safely be ignored for the sake of clarity and without loss of generality in the current context. Therefore, it is reasonable to categorise mobility in the urban context into macroscopic, mesoscopic, and microscopic spatiotemporal scales. From a transport systems perspective, travel patterns including trip origins, destinations, schedules, and travel modes form a relatively macroscopic analysis (on the left-hand side in Figure 1). Transport demand models at this level characterise activity-travel patterns across transport zones. Once an individual has arrived at the zone-level destination, modelling which specific building within the zone the individual will choose and when the individual will move between buildings within the given transport zone is a mesoscopic problem. Within the chosen building, modelling indoor occupancy and transitions is the microscopic element of the framework (on the right-hand side of Figure 1); this has traditionally been the focus of research in occupant-related building performance studies.

Figure 1.

A conceptual framework for integrating activity-travel behaviour and building occupancy models.

Establishing a link between building and transport systems in this manner provides opportunities for generating novel insights into these two domains. A key component in this framework is the mesoscopic model that can translate between the macro perspective of the transport system and the micro perspective of the building. For example, the monitored occupancy patterns for multiple buildings within an area are valuable inputs to infer real-time origin-destination (OD) trips for transport planning and management. Moreover, analysis of local travel patterns could also benefit from sensor-detected occupancy patterns (e.g. Wi-Fi or CCTV cameras) in a group of neighbouring buildings. From a building management perspective, arrival and departure patterns derived from OD trip data can support a more realistic initialisation of building occupancy simulation. Furthermore, the mesoscopic occupancy model that describes occupancy patterns of multiple interconnected buildings can play a role in understanding and planning energy and power needs in a district, which is a crucial consideration in the context of distributed and local energy generation and storage.

As for the implementation, either coarse (e.g. travel survey data) or fine (e.g. GPS tracking data) inputs can feed the model to generate activity-travel patterns (see Figure 2). Arrival distributions inferred from these patterns are used to initialise the first arrival¹ states for the building occupancy simulation. When linking travel and occupancy models at the mesoscale, an interface is required to reconcile arrival and departure outputs between the two models unless fine-grained travel data that contains building-specific locations are available.

Figure 2.

The workflow of linking transport parameters to building modelling.

Our framework builds upon existing literature while presenting methodological differences. Unlike Mosteiro-Romero et al. (2020), who adapted a transport simulator (i.e. MATSim) to scale up building occupancy modelling to the mesoscopic level, or Rezvany et al. (2021, 2023), who focused on the macroscopic travel-activity problem, our framework uniquely bridges travel and building occupancy models by systematically reconciling their respective inputs and outputs via the proposed mesoscopic link (Figure 1). We also integrated urban covariates, such as operational conditions of surrounding underground stations, work schedule policies, and external air temperature, into the modelling framework to capture the complex interactions. This integration enables explicit modelling of the dynamic interactions between transport systems and building occupancy patterns. Rather than developing a computationally demanding high-resolution model, our approach offers adaptability through its novel combination of established macro- and micro-scale modelling techniques. Such a flexible coupling mechanism enables our framework to be extended for different buildings/districts with their models and inputs/outputs exchanged via the standardised format.

Operational framework for linking an activity-travel demand simulator and a hazard-based building occupancy model

The first arrivals at buildings should naturally depend on travel behaviour ahead of the arrivals. To link behaviour in urban transport and building systems, this section presents a systems approach to operationalise the framework, accounting for interactions between the two systems as proposed previously (Figure 1). In particular, we propose the modelling workflow in Figure 2 to link simulations of travel and building occupancy behaviours.

Macroscopic component of the framework

In practice, an ABM of travel demand, developed by Pawlak et al. (2021, 2022), was used to generate individuals’ travel and activity schedules (or plans). In their study, Pawlak et al. (2021, 2022) developed the ABM using UK TUS data to generate simulations of household energy demand. In this study, the ABM was incorporated as the macroscopic part of the implementation framework to simulate the individuals’ in-home and out-of-home activities (work, study, etc.) and their temporal sequencing during the day. A limitation of the TUS data is that it does not include information on specific buildings where activities took place. Therefore, we assumed that the activity-start times generated in the ABM represent the times of first arrival at buildings which accommodate the corresponding activity types.

Mesoscopic component of the framework

Given arrival times at multiple buildings within a zone (meso-scale), modelling which building an individual would arrive at is a destination choice problem. The conventional method to address such a problem is to use discrete choice models that apply utility functions to characterise individuals’ preferences between a set of possible locations, in which the effects of locational attributes and exogenous factors can be estimated (Fotheringham et al., 2001). However, research in the building context focuses on modelling arrival states with respect to not only the spatial dimension but also accounting for temporal variation. Therefore, we propose to use a two-parameter log-logistic distribution to describe the variation of individuals’ preferences for each destination against the time of day, due to its flexibility and efficiency in capturing various patterns (e.g. involving domestic and non-domestic buildings).

The building-specific distribution is intended to describe the duration between midnight (00:00) and the time of an individual’s first arrival at a building. Mathematically, the log-logistic form is controlled by a shape parameter $κ$ (>0) and a scale parameter $η$ (>0). The distribution is unimodally shaped when $κ$ >1; it has a monotonically decreasing shape when $κ \leq 1$ . The probability density function (PDF) of the log-logistic distribution is given by

f (t; η, κ) = \frac{(κ / η) {(t / η)}^{κ - 1}}{{(1 + {(t / η)}^{κ})}^{2}}

(1)

These parameters ( $κ$ and $η$ ) can be expressed as the exponential of a linear regression to ensure the positivity of the estimated parameters. Explanatory variables $X$ were included in the scale parameter to characterise exogenous impacts on arrival distribution. When comparing empirical arrival patterns (shown in Supplemental Material S3), the shape of the arrival time distribution was found to be driven by activity purpose and building type rather than external surroundings and hence only building-specific constant terms were included in the shape parameter. To capture differences between the same type of buildings within a zone, exogenous covariates were specified to capture surrounding effects on the scale of the arrival time distribution. Thus, these parameters are given by

η = \exp (a_{0} + a_{1} X_{1} + \dots)

(2)

κ = \exp (k_{0})

(3)

Using Bayes’ theorem, the conditional probability of an individual’s arrival at building b at time t, given that the individual has arrived in the area at time t, is defined by

P (b | t) = \frac{f_{b} (t; η_{b}, κ_{b})}{\sum_{g = 1}^{B} f_{g} (t; η_{g}, κ_{g})}

(4)

where

f_{b}

is the PDF specified for building b within the area;

η_{b}

and

κ_{b}

are building-specific parameters in the PDF. The conditional probability distribution expresses the probability of arrival at building b relative to arrivals at all possible buildings in the same area at the same time-point t.

The parameters of the first arrival distributions were estimated using the maximum likelihood method, given by

L (η_{b}, κ_{b} | t, X) = \prod_{i = 1}^{N} \prod_{b = 1}^{B} {(f_{i, b} (t_{i, b}; η_{b}, κ_{b}))}^{δ_{i, b}}

(5)

where N is the total number of individuals observed in the sample and B is the total number of buildings within the area;

δ_{i, b}

is a dummy indicating the building b is chosen by individual i at their first arrivals.

Microscopic component of the framework

The microscopic end (i.e. building occupancy simulation) was realised using a nested hazard-based model (NHBM) developed by Hou et al. (2022). The NHBM not only uses the generated first arrival states but also incorporates transport impact on intermediate occupancy and circulation across multiple buildings. Theoretically, NHBM is an individual-level microsimulation that models occupancy sequences both temporally, from an individual’s arrival to departure, and spatially, across indoor spaces within multiple buildings. It can describe an individual’s occupancy durations and location choices with the impacts of exogenous urban variables (operations of surrounding public transport network, work schedule policies, external air temperature, etc.) reflecting the dynamic surroundings. In the NHBM simulation, occupancy durations and location choices are determined through random sampling from the distributions calibrated using detected occupancy and transition data. Finally, the microsimulation produces a series of occupancy states for each individual and every day (i.e. when an individual arrives at a specific building, when and where they move inside it, and when they leave it). Aggregating the simulation outputs produces the occupancy-count patterns within each building, which is the kind of information typically used to aid with building and facility energy controls. Further details of the macroscopic and microscopic models are presented in the Supplemental Material.

Case study

Case description

The South Kensington (SK) campus of Imperial College London is the case study area for which the framework was implemented and demonstrated. It is located in central London, UK, and is well connected with the transport infrastructure, making it a good case study for demonstrating the proposed framework. Two underground metro stations close to the SK campus, within 15-min walking distance, provide three underground lines with high traffic for outbound and inbound travel. The reasonable walking distance to the stations and high passenger demand ensure that the population on campus are highly likely to choose this travel mode. Accordingly, incidents on these underground lines would cause significant impacts on arrivals and building occupancy within the campus.

The SK campus includes 35 well-connected buildings with multiple functions (cf. Figure 3), where a variety of events drive frequent circulation of occupants between buildings. The physical connections between the buildings support the high demand for circulations of the occupants, serving as a good example of interdependent buildings in high-density areas across London that urgently require updates to their energy grids.

Figure 3.

Spatial distribution of multi-type buildings and their aggregation in this study.

To facilitate the operationalisation of our modelling framework, these buildings were aggregated into 10 groups according to the layout and the population they serve (Figure 3). Amongst the buildings on campus, the library was the only one open 24 h and was hence assigned to a separate group, despite serving the same type of population as teaching and office buildings. The faculty building is an administrative building that only permits limited staff access and requires authorisation, setting it apart as a ‘closed space’ compared to the other groups.

This study used anonymous Wi-Fi logs to infer individuals’ transitions across more than 250 access points (APs) on campus. Table 1 shows the format and information contained in the Wi-Fi dataset, in which location (indicated by AP name) and time of accessing Wi-Fi for each user were recorded. Although wireless clients re-authorise every 30 min as per the default setting, more frequent re-authorisations occur under certain situations, especially when roaming between Wi-Fi AP groups. The roaming between multiple APs (in Table 1) reflects the user’s physical movements, whilst updates due to the default scan indicate that the user stays in one location.

Table 1.

The format of Wi-Fi probing logs that shows one user’s trajectories.

Wi-Fi authentication logs
Unix time	User category	Hashed user ID	Hashed MAC address	Building ID	Floor ID	Network name
1635386400	Student	001	6316b0xxxx	me	3	Imperial-WPA	When the user moves
1635387000	Student	001	6316b0xxxx	me	1	Imperial-WPA	When the user moves
1635387300	Student	001	6316b0xxxx	fac	1	Imperial-WPA	When the user stays for more than 30 min
1635389100	Student	001	6316b0xxxx	fac	1	Imperial-WPA	When the user stays for more than 30 min
…	…	…	…	…	…	…

The data for the case study was collected during October and November 2021 when the campus first reopened for teaching activities after the COVID-19 pandemic. During the study period, the occupancy count on campus, derived from the Wi-Fi data, varied between 10,000 and 13,000 individuals on workdays. The global peak of the first arrival states on campus is observed at 9:00, whereas a slight increase of first arrival is seen at 14:00. The campus population consists of five categories: undergraduate (UG) students (45.6%), postgraduate (PG) taught students (13.6%), PG research students (11.4%), staff (14.0%) and others (15.4%; including visitors, guests, alumni, etc.). Staff and PG research students have either individual or shared offices and hence their stays on campus are mainly based in their offices. In contrast, the locations of UG and PG taught students and others are driven by the arrangement of courses and events.

Model implementation

This study demonstrates the novel connection between macroscale travel and microscale building occupancy models via a mesoscopic model that enables cross-scale data transfer and system interaction analysis. The mesoscopic model was calibrated using Wi-Fi connection logs, where initial daily connections for each individual were interpreted as arrival states (220,886 observations during the study period). The model was specified with urban covariates as eq. (2), including the time of day, building type, and binary variables indicating the occurrences of the nearest metro station closure. These variables account for exogenous influences on arrival distributions amongst campus buildings. Parameter estimation using the maximum likelihood method was realised using the ‘optimx’ package in the R programming language, with the log-likelihood function as given in eq. (5). The Chi-square goodness of fit test was used to evaluate model calibrations.

The mesoscopic simulation was initialised using the outputs of the ABM simulator with UK TUS data as feeds, where individual arrival times were determined as a function of key socioeconomic variables (e.g. car ownership). This approach can capture how transport factors influence personal daily activity-travel patterns. The derived activity agendas, including activity sequences and timing information, were then used to derive arrival times for individuals entering the case study area. Subsequently, the calibrated mesoscopic model assigned these arrival times to specific campus buildings, using a Monte Carlo approach and probabilities calculated as eq. (4). The number of individuals with different roles (students vs staff) in the simulation was determined according to their empirical shares in the population. The mesoscopic simulation finally produced individual-level arrival profiles for each building, which form the inputs for the subsequent building occupancy simulation. For occupancy simulation, NHBM adopted in this study was calibrated using the same Wi-Fi dataset that captured some occasions of transport impacts on occupancy sequences (e.g. tube delays and strikes). The NHBM simulation, incorporating transport covariates and arrival inputs, subsequently created occupancy and transition patterns among multiple buildings.

Joint transport and building occupancy modelling scenario

In this section, we present a scenario simulation to specifically showcase how a transport incident could impact both first arrivals and intermediate occupancy and transitions between buildings in an area. The scenario emulates a situation when the underground station nearest to the SK campus is closed, and we see that the joint model captures the impacts we anticipate in terms of longer travel times for most users (due to changes to travel route and mode) and corresponding delays to occupants’ arrivals at buildings. The following analysis draws upon data from two different sources (UK TUS and Imperial College Wi-Fi data), hence the results carry only limited empirical representativeness and serve more to demonstrate the operational principles of the framework.

Estimation results of the mesoscopic model

Distributions of the first arrival times across multiple buildings were computed using the proposed log-logistic model, calibrated with the feeds of Wi-Fi logs. The estimation results of the model are illustrated in Table 2, which show that the specified covariates do well in capturing the distinctive arrival patterns across buildings. The plots in Figure 4 compare the modelled and observed probabilities of first arrival for the different building groups. Overall, the Chi-square goodness of fit test indicates no statistically significant differences between the observed and modelled distributions (p-value >.05).

Table 2.

Estimated parameters of the log-logistic model for arrival distribution.

Variable	Group 4	Group 7	Sherfield	Library	Faculty	Other buildings
Variable	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)	Coeff. (SE)
Constant $a_{0}$	2.960* (0.002)	2.777* (0.006)	2.920* (0.002)	2.903* (0.002)	2.901* (0.015)	2.908* (0.001)
5:00–7:00	−1.144* (0.008)	−0.976* (0.013)	−1.047* (0.004)	−1.031* (0.008)	−1.064* (0.033)	−1.039* (0.002)
7:00–9:00	−0.826* (0.003)	−0.679* (0.007)	−0.790* (0.002)	−0.747* (0.002)	−0.754* (0.015)	−0.742* (0.001)
9:00–11:00	−0.675* (0.003)	−0.488* (0.007)	−0.621* (0.002)	−0.608* (0.002)	−0.635* (0.015)	−0.623* (0.001)
11:00–13:00	−0.475* (0.003)	−0.292* (0.008)	−0.430* (0.002)	−0.423* (0.002)	−0.437* (0.016)	−0.433* (0.001)
13:00–15:00	−0.335* (0.003)	−0.149* (0.008)	−0.295* (0.002)	−0.272* (0.002)	−0.265* (0.016)	−0.275* (0.001)
15:00–17:00	−0.190* (0.003)	—	−0.163* (0.002)	−0.138* (0.002)	−0.144* (0.018)	−0.144* (0.001)
Transport incident	−0.050* (0.001)	−0.050* (0.001)	−0.050* (0.001)	−0.050* (0.001)	−0.050* (0.001)	−0.050* (0.001)
Goodness of fit
The number of observations: 220,886
Log-likelihood at convergence: −1341104
Log-likelihood at parameters constrained to zero: −3576134

*p-value $\leq 0.05$ .

Figure 4.

Log-logistic model fitting for arrival distributions over buildings compared to observed plots.

Building occupancy patterns driven by transport incidents

In order to simulate the transport incident scenario, it was assumed that people with cars would arrive at the destination buildings on time while others without cars would experience a 30-min or 1-h delay of first arrival. This reflects the fact that the latter group would have to change their public transport route or travel mode due to the sudden closure of the nearest underground station. An empirical study by Yin et al. (2016) provides compelling evidence for our assumption that over 25 min of station closure leads to travel changes for all passengers. The observations from our Wi-Fi logs that capture the station closure due to a tube strike suggested the longer travel time and the postponed campus arrivals due to changes in travel plans (in Supplemental Material S4).

Taking into account these delays as a function of car ownership in the macroscopic ABM, the resulting arrival times were fed into the mesoscopic model and distributed across buildings. First arrival states for each building were thus established and used to initiate the subsequent NHBM building occupancy simulation. In the NHBM simulator, covariates representing transport incidents besides the affected arrival times were included in the occupancy model and reflected as the states occurring across the day. With all parameters fed into the workflow, the NHBM ultimately produced building occupancy patterns under the postulated transport scenario (in Figure 5), calculated using the mean values of 100 simulations for each group.

Figure 5.

Simulations of occupancy count against the time of day for do-nothing scenario, 30 min and 1 h delay for individuals without cars.

In terms of predicted patterns, building groups without residences (i.e. excluding Groups 4 and 7) show the expected shifts in arrival to later times, while Groups 5 and 6 and the faculty building exhibited varied shapes of occupancy patterns. This result is consistent with findings in the literature that individuals would postpone their journey (Monsuur et al., 2021) or change their travel mode or route (Yin et al., 2016), resulting in longer travel times (e.g. 90 min taken for bus replacement in Currie and Muir (2017)). This is also supported by our observations of the delayed arrivals (see Supplemental Material S4). Notably, the results indicated almost identical patterns under different scenarios for Groups 4 and 7 with a high proportion of residential space. This is not surprising given that the residents are mainly UG students who do not need to use the public transport network and are insensitive to transport incidents. A comparison between the patterns of Group 7 and other groups reflects the nearly constant variations across the day, which could stem from the low transitions out of the residential buildings (i.e. occupancy rate drops by only 0.03) during working hours. The reason was confirmed by the inspections undertaken by the College’s building managers, who stated that students took online lectures in the daytime within their residences and the occupancy demand was high due to the COVID-driven restrictions to physical attendances during the case study period. In addition, slight increases in occupancy between 7:00 and 9:00 in Group 7 reflect staff coming to the office spaces in this group of buildings.

In contrast, the faculty building, only authorised to access by administrative staff, was much more sensitive to transport incidents, reflected in significant occupancy changes. This result could be due to the more flexible working agenda of people in the faculty building, compared to inflexible teaching schedules which are the dominant pattern in most other buildings. The model also predicts significant delays of arrivals in the morning for the faculty building, with considerable changes in the shape of occupancy patterns when a severe delay (1 h delay) occurs. The predicted patterns suggest that minor impacts (i.e. 30 min delay) generally delay individuals’ arrival times in the morning, without changes in the afternoon. When the impact was more severe, delays of morning arrivals were cumulated between 7:00 and 10:00. Subsequently, more outbound transitions happened in the afternoon, possibly stemming from the significant changes in the work-activity sequences (reflected as the occupancy sequences in the NHBM outputs).

Variations in local transition patterns

The simulations can also produce a local-OD matrix within the campus for any time interval, akin to OD matrices used for facility planning at the district level. Figure 6 presents the OD matrices of transitions between buildings generated for do-nothing and postulated transport scenarios during 7:00–12:00, 12:00–14:00 and 14:00–17:00, with the colours representing the (ranges in) number of people transitioning between building groups. Higher transitions reflect higher demand in corridors or pathways between buildings and vice versa. Generally, higher transitions appear for inbound and outbound flows of Group 1, which is as expected since Group 1 contains three large-scale teaching buildings around the main entrance of the campus. Group 1 is also located where it is well connected to all the other building groups and high pedestrian flow into/out of Group 1 buildings is expected due to the frequent organisation of internal and external events. In addition, significantly higher transitions inbound and outbound of the Sherfield building were observed from lunchtime onwards due to its restaurant and food offers.

Figure 6.

OD matrices between building groups for do-nothing, 30-min-late and 1-h-late scenarios during 7:00–12:00, 12:00–14:00 and 14:00–17:00.

The model results suggest that transport disruptions could also impact individuals’ transitions between the buildings. During lunchtime (12:00–14:00), transitions between buildings significantly increase under both scenarios of transport incidents, with no significant differences across the two scenarios. In the afternoon (14:00–17:00), the results indicate no impacts on transitions related to residential spaces (Groups 4 and 7). In contrast, building groups, containing teaching and office activities, exhibit significantly increased outbound and inbound transitions in the afternoon. These findings aid in understanding flows across the campus when affected by transport incidents, which in turn can impact energy demand and use patterns. Such details cannot be identified only from occupancy patterns, especially when overall occupant counts remain static (Figure 5).

Discussions

Potential applications

The results show how the proposed framework captures the effects of transport incidents on first-arrival states, intermediate occupancy and transitions across buildings within a campus. Such results could be translated into actionable insights, contributing to data-driven energy management at the district-level smart grids. In this study, the results of the empirical application provide evidence-based information for decision-making in planning and managing districts with a similar type and scope.

For district planners, the model enables the estimation of building electricity demands, as associated with occupancy counts, for different transport scenarios. This study demonstrates the impact of arrival delays due to transport incidents. Specifically, we find that the occupancy-induced electricity demand of buildings with closed space (i.e. strict permissions for access) will be sensitive to surrounding transport incidents. In contrast, the occupancy-induced demand of buildings with free access is likely to remain unchanged overall across different transport scenarios, despite slight changes in the morning patterns. Therefore, closed buildings and spaces could benefit from flexible and demand-driven control of electricity supply (e.g. smart grids with energy storage). In addition, the proposed modelling framework can be used to anticipate and understand how people transition between buildings within a district. For instance, we found that arrival delays due to transport incidents generally led to decreased transitions during the morning but an increase during lunchtime and the afternoon. This warns district planners to expect increased energy demand variability (instability) from noon onwards, especially for buildings with more physical connections to others. Planners could accordingly manage the power supply and/or install smart devices and grids for buildings with high circulation.

For building estate managers, the predicted occupancy patterns help design time schedules for running HVAC systems. For example, the shifts in arrival times due to transport disruptions suggest reasonable adjustments for preconditioning and setback periods in the case study buildings to coordinate with the events in the transport system and achieve higher building energy efficiency. It must be noted, however, that the analysis in Results Section is merely illustrative. In real-world applications, monitored (sensor-based) travel data for specific buildings would be required to produce realistic predictions of travel patterns as they impact HVAC control strategies.

For facility managers, occupancy counts within buildings can reflect the demand for surrounding micromobility (e.g. bikeshare and e-scooters) and electric charging facilities. The big changes in occupancy-driven demand (induced by transport incidents) in closed buildings, and ones with physical connections to other buildings, suggest that facility managers could manage the parking and charging equipment more efficiently to meet the dynamic demand. Additionally, transitions between buildings can shed light on the uses of corridors between buildings and elevators within a building, which can help facility managers improve signage and design of spaces. Such information would also be useful in developing fire evacuation plans for the buildings when dealing with atypical situations such as transport incidents.

Limitations

Although the framework proposed in this study has been operationalised to analyse district-level energy-related behaviours, there remain limitations to be addressed in order to enable its applications in broader and more complex urban contexts.

Implementation of the integrated framework ideally involves individual-level microsimulation, using data that can record individual trajectories. In this study, Wi-Fi data was the sole input to calibrate the mesoscopic model, based on an assumption of high penetration of Wi-Fi technologies and mandated data sharing. The Wi-Fi data was anonymised and collected with the permissions of the College’s Data Protection Policy, following the General Data Protection Regulation (GDPR). Due to the private information in traceable data (including smart card and CCTV camera data, etc.), it cannot be shared on any open data portals and can only be revealed at an aggregate level. Overall, two approaches can address the challenges of data access in further model adaptions (e.g. in the contexts of residential and business districts). If the detailed data at a user level (e.g. CCTV) is employed for modelling, it is necessary to combine data-sharing agreements (e.g. GDPR) and anonymisation techniques to address privacy concerns. It is also worth exploring the use of incentives for individuals to share such data, either in the form of money or a service. Alternatively, reconstructing the individual trajectories based on partial records, akin to OD matrix estimation using traffic counts (e.g. GPS and mobile phones), can also help to operationalise the conceptual framework, though the modelling procedure will require further modifications.

Although the results of the analysis presented in this paper are specific to the empirical context, the model implementation has demonstrated the operational principles of the framework. However, scaling the model to larger urban settings with more complex transport networks requires further research. For example, the framework represents bi-directional interactions between building and transport systems, but this empirical study only considers the direction of impact from the transport system to the building system. The reverse effect, such as a scenario of where a building closure caused by flooding would impact the travel demand patterns, calls for future investigation. The college campus in our case study is a very small part of the wider and more populated London neighbourhood, and as such will have a very small impact on the transport system. A larger-scale case study should be undertaken to better understand the complexities. For example, in the context of vehicle electrification, integrating building occupancy simulation and travel demand simulation would help to optimise electric vehicle (EV) charging schedule and management (e.g. energy supply and storage) for charging stations around the city. Naturally, building occupancy profiles to some extent can reflect the use of the surrounding EV charging stations (through charging duration profiles).

Conclusions

Interactions between transport and building systems are often encountered in urban areas but have remained underexplored in past transport and building occupancy modelling studies. This paper proposes to integrate the modelling of travel and building occupancy demand whilst considering the connections between an individual’s mobility behaviour and their occupancy behaviour. A new systems approach to the framework is presented to elaborate the links between the two domains conceptually and as a means for combining the models operationally. Following the proposed modelling principles, an empirical application for the South Kensington campus of Imperial College London, UK, was developed, which demonstrates the use of the proposed framework and its benefits to enhancing data-driven energy management for distributed and smart grids, with the high penetration of electrification strategies in transport and building systems. In future, electrification strategies further strengthen the connections between transport and buildings (e.g. EV charging points surrounding the buildings), and increasing computational capabilities and data availability will accelerate the integration of modelling multiple systems within the urban context. The proposed framework and systems approach provide a simulation tool for future scenarios when facing uncertainties. For example, with natural hazards due to climate change, the sudden changes in people’s travel and activity patterns would increase the instability of energy demand. In addition, any social distancing and isolation policies in pandemic-like scenarios would result in shifts of energy demand from transport networks to buildings. The proposed integrated framework, through capturing the interactions between users and interdependent infrastructures in their simulations, would help to enhance the planning and management of energy facilities to confront future uncertainties (e.g. efficiently investing in smart grid facilities).

Supplemental Material

Supplemental Material - Integrating activity-based transport and building occupancy models for campus-scale energy management

Supplemental Material for Integrating activity-based transport and building occupancy models for campus-scale energy management by Huiqiao Hou, Jacek Pawlak and Aruna Sivakumar in Journal of Environment and Planning B: Urban Analytics and City Science.

Footnotes

Author contributions

Huiqiao Hou: Conceptualisation, Methodology, Formal Analysis, Data Curation, Writing – Original Draft, Visualisation; Jacek Pawlak: Writing – Review & Editing, Supervision; Aruna Sivakumar: Writing – Review & Editing, Supervision.

Declaration of conflicting interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The first author was financially supported by the China Scholarship Council and Imperial College London (Grant No. 201808230110). The second and third authors acknowledge support from the Integrated Development of Low-Carbon Energy Systems (IDLES) research programme at Imperial College London funded by the Engineering and Physical Sciences Research Council (EPSRC) Programme Grants (EPSRC Grant No. EP/R045518/1).

ORCID iDs

Huiqiao Hou

Jacek Pawlak

Aruna Sivakumar

Data Availability Statement

Data access requests can be made to the authors.*

Supplemental Material

Supplemental material for this article is available online.

Note

Huiqiao (Heather) Hou is a Research Associate at the School of Urban Planning and Design, Peking University. She obtained her PhD and MSc degrees from Imperial College London, financially supported by the China Scholarship Council and Imperial College London. With an academic background in Civil Engineering, Transport Modelling, and Building Energy, she specialises in the interdisciplinary integration of transport and building systems. Her research focuses on leveraging emerging data sources (e.g. Wi-Fi connections and traffic APIs) to model human behaviours, supporting the planning and management of energy systems across scales, from a single building to urban districts.

Jacek Pawlak is a Research Fellow at Imperial College London (Urban Systems Lab and Centre for Transport Engineering and Modelling) working in the fields of transport, urban systems, technology, and innovation. His background is in Economics, Geography and Transport Studies (Imperial College London and University College London). Since January 2017 he has been a Committee Research Coordinator for the US Transportation Research Board Standing Committee on Effects of ICT on Travel Choices (AEP35). He is also a Fellow of the Royal Geographical Society. Dr Pawlak also collaborates with Cisco Innovation for the UK & as a Principal Modelling Specialist.

Aruna Sivakumar is a Professor of Consumer Demand Modelling & Urban Systems at the Centre for Transport Engineering and Modelling, Imperial College London. She is the director of Urban Systems Lab, and leads several smart city and systems modelling initiatives including the EU Sharing Cities project, decentralised energy demand modelling in the EPSRC-funded IDLES project, accessibility framework for equity analysis in the Wellcome Trust-funded Pathways project. Over a 15-year career, Prof. Sivakumar has published over 40 peer-reviewed journal papers and several book chapters. Her papers on urban energy systems and activity-based travel demand modelling have both been cited over 400 times.

References

Axhausen

Horni

Nagel

(2016) The Multi-Agent Transport Simulation MATSim. London, UK: Ubiquity Press.

Bon

den Duijn

Dukai

, et al. (2016) Identifying movement patterns from large scale wifi-based location data: a case study of the TU Delft campus. Geomatics Synthesis Group Project. http://resolver.tudelft.nl/uuid:bd426086-5ccb-4e3e-9472-b6f4af6a9956.

Chen

Soh

(2015) Modeling regular occupancy in commercial buildings using stochastic models. Energy and Buildings 103: 216–223.

Currie

Muir

(2017) Understanding passenger perceptions and behaviors during unplanned rail disruptions. Transportation Research Procedia 25: 4392–4402.

De Palma

Rochat

(1999) Understanding individual travel decisions: results from a commuters survey in Geneva. Transportation 26(3): 263–281.

D’Oca

Hong

(2015) Occupancy schedules learning process through a data mining framework. Energy and Buildings 88: 395–408.

Fotheringham

Nakaya

Yano

, et al. (2001) Hierarchical destination choice and spatial interaction modelling: a simulation experiment. Environment and Planning A: Economy and Space 33(5): 901–920.

Giuliano

(2021) Analyzing traffic impacts of planned major events. Transportation Research Record: Journal of the Transportation Research Board 2675(8): 432–442.

Griffioen

Vermeer

Dukai

, et al. (2017) Exploring indoor movement patterns through eduroam connected wireless devices. In: Proceedings of the 20th AGILE international conference on geographic information science, Wageningen, The Netherlands, 9–12 May 2017. Wageningen University.

10.

Happle

Fonseca

Schlueter

(2020) Context-specific urban occupancy modeling using location-based services data. Building and Environment 175: 106803.

11.

Hickman

Banister

(2014) Transport, Climate Change and the City. London, UK: Routledge.

12.

Hou

Pawlak

Sivakumar

, et al. (2022) Building occupancy modelling at the district level: a combined copula-nested hazard-based approach. Building and Environment 225: 109661.

13.

IEA (2021) Key World Energy Statistics 2021. Paris, France: IEA. License: CC BY 4.0. https://www.iea.org/reports/key-world-energy-statistics-2021.

14.

IEA (2022) Transport. Paris, France: IEA. License: CC BY 4.0. https://www.iea.org/reports/transport.

15.

Kalogianni

Sileryte

Lam

, et al. (2015) Passive WiFi monitoring of the rhythm of the campus. In: AGILE 2015, Lisbon, Portugal, 9–12 June 2015, pp. 9–14.

16.

Kwoczek

Di Martino

Nejdl

(2014) Predicting and visualizing traffic congestion in the presence of planned special events. Journal of Visual Languages & Computing 25(6): 973–980.

17.

Monsuur

Enoch

Quddus

, et al. (2021) Modelling the impact of rail delays on passenger satisfaction. Transportation Research Part A: Policy and Practice 152: 19–35.

18.

Mosteiro-Romero

Hischier

Fonseca

, et al. (2020) A novel population-based occupancy modeling approach for district-scale simulations compared to standard-based methods. Building and Environment 181: 107084.

19.

Page

Robinson

Morel

, et al. (2008) A generalised stochastic model for the simulation of occupant presence. Energy and Buildings 40(2): 83–98.

20.

Pawlak

Faghih Imani

Sivakumar

(2021) How do household activities drive electricity demand? Applying activity-based modelling in the context of the United Kingdom. Energy Research & Social Science 82: 102318.

21.

Pawlak

Imani

Sivakumar

(2022) Scenario-based modelling of residential energy demand uncertainty using activity-based approach. In: Mapping the energy future-voyage in uncharted territory-43rd IAEE international conference, Tokyo, Japan, 31 July 2022–3 August 2022. International Association for Energy Economics.

22.

Rakha

Rose

Reinhart

(2014) A framework for modeling occupancy schedules and local trips based on activity based surveys. In 2014 building simulation conference, Atlanta, GA, 10–12 September 2014, pp. 433.

23.

Rezvany

Hillel

Bierlaire

(2021) Integrated models of transport and energy demand: a literature review and framework. In Proceedings of the 21st Swiss transport research conference (STRC), Ascona, Switzerland, 12–14 September 2021.

24.

Rezvany

Bierlaire

Hillel

(2023) Simulating intra-household interactions for in- and out-of-home activity scheduling. Transportation Research Part C: Emerging Technologies 157: 104362.

25.

Salim

Dong

Ouf

, et al. (2020) Modelling urban-scale occupant behaviour, mobility, and energy in buildings: a survey. Building and Environment 183: 106964.

26.

Traunmueller

Johnson

Malik

, et al. (2018) Digital footprints: using WiFi probe and locational data to analyze human mobility trajectories in cities. Computers, Environment and Urban Systems 72: 4–12.

27.

Wang

Shi

, et al. (2020) Dynamic occupant density models of commercial buildings for urban energy simulation. Building and Environment 169: 106549.

28.

Wieczorek

Rohracher

Bauknecht

, et al. (2024) Citizen-led decentralised energy futures: emerging rationales of energy system organisation. Energy Research & Social Science 113: 103557.

29.

Yan

Hong

Dong

, et al. (2017) IEA EBC annex 66: definition and simulation of occupant behavior in buildings. Energy and Buildings 156: 258–270.

30.

Yin

Han

, et al. (2016) Modeling and simulating passenger behavior for a station closure in a rail transit network. PLoS One 11(12): e0167126.

31.

Zakaria

Trivedi

Cecchet

, et al. (2022) Analyzing the impact of COVID-19 control policies on campus occupancy and mobility via WiFi sensing. ACM Transactions on Spatial Algorithms and Systems (TSAS) 8(3): 1–26.

32.

Zhou

Yang

(2016) Big data driven smart energy management: from big data to big insights. Renewable and Sustainable Energy Reviews 56: 215–225.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.38 MB