People flow in buildings

Accurate modeling of how people use elevators and behave in emergencies is the key to successful people flow planning.

Modern elevators contain many sensors and software designed to move people safely and comfortably inside the buildings.

Modern elevators contain many sensors and software designed to move people safely and comfortably inside the buildings. Photo courtesy of 123rf.com | © Inspirestock International

By Juha-Matti Kuusinen, Janne Sorsa, Marja-Liisa Siikonen, Henri Hakonen and Harri Ehtamo

In 2014, more than half of the total global population was living in an urban environment. In the urban world, elevators often form the core of the buildings, whose value is thus largely dependent on their performance and the vertical transportation system as a whole. In addition to smooth and efficient people flow, the quality of a building depends on safety, especially on fast evacuation in an emergency situation.

This article describes some key advances in people flow and building safety modeling based on several years of collaborative research carried out by KONE Corporation, one of the global leaders in the elevator and escalator industry, and Systems Analysis Laboratory of the Aalto University. This ongoing research forms the basis for the development of efficient vertical transportation systems, as well as advanced people flow and evacuation planning tools which provide important decision support for building designers.

Elevators Adapt to People Flow

Modern elevators contain many sensors and software designed to move people safely and comfortably inside the buildings. The elevators are typically combined into groups, and the elevators in the same group are controlled by an elevator group control system (EGCS). The main task of the EGCS is to dispatch the elevators to passengers’ calls. To transport passengers to their destinations fast and without waiting, modern EGCS use advanced mathematical optimization algorithms and passenger traffic forecasts to adapt to the possible changes and uncertainties in the traffic [16]. The EGCS measures the people flow in a building by counting the number of passengers boarding and alighting at each elevator stop and passengers’ calls. The measurements are made for each floor and direction throughout the day [10]. The EGCS forms the traffic forecasts from the measurements and uses them in dispatching the elevators to the calls in an efficient way.

The floor- and direction-based statistics, however, do not fully describe the passenger traffic in a building, and thus cannot be used to explicitly predict all the important uncertainties related to the traffic. For example, it is not possible to accurately predict how many passengers typically travel between a given origin-destination floor pair, i.e., how much free space is typically required in an elevator during a journey from the origin to the destination. The prediction of this and other important information related to the passenger traffic requires the measuring of every passenger journey between every pair of floors in a building. This, however, is not possible with the commonly used sensors.

A recent study presents a method that can be used to estimate the passenger journeys, and thus, to improve the performance of the elevators [9]. The passenger journeys can also be combined into a building origin-destination matrix to recognize the three traditional traffic components – incoming, outgoing and interfloor – and special floors such as express zones. The three traffic components together with the total traffic intensity define a range of traffic patterns, for example, up-peak traffic during which most people travel from the main elevator lobby or ground floor to the populated floors.

Figure 1: A building origin-destination matrix based on the passenger journeys estimated during a 15-minute simulation of mixed lunch hour traffic in a 25-floor office building

Figure 1: A building origin-destination matrix based on the passenger journeys estimated during a 15-minute simulation of mixed lunch hour traffic in a 25-floor office building. Source: Kone Corporation

Sociality Completes Traffic Planning

No matter how intelligent the elevator group control is, the passenger traffic cannot be efficiently handled if, for example, there are too few elevators in the group. On the other hand, elevators are expensive and take a lot of floor space, and thus, too many elevators would also be a bad solution, especially for the building owner. For a long time, elevator traffic planning has been based on the theoretical up-peak roundtrip model, which is used to calculate the up-peak performance of an elevator group [14]. In cases involving advanced group controls or unusual building configurations, the planning is typically based on simulation. The results of the traffic simulations depend on the simulated traffic pattern and on the assumptions of how passengers behave and use elevators. Hence, to be able to plan appropriately dimensioned elevator groups where the number and size of the elevators is not too large or small, the simulated behavior should model the real behavior as accurately as possible.

One of the key inputs in elevator traffic simulations is the process that describes how passengers arrive at the elevator lobbies and register calls. The typical assumption is that passengers arrive according to a Poisson process. In this process, passengers arrive individually with exponentially distributed inter-arrival times. It has been shown that this assumption holds in the real world during an up-peak traffic situation [1]. Kuusinen et al. [8] present a study on the arrival process in a high-rise office building, taking into account, for the first time, that the process does not remain the same throughout the day. For example, during lunch time, people socialize and move in batches whereas in the morning they arrive at work alone. Hence, the real arrival process can be modeled more accurately with a compound Poisson process where the passengers arrive in batches regardless of the time of day. Of course, the size of the arriving batches need to be defined, which can be done with a batch-size distribution.

The batch-size distribution has been measured in several office buildings around the world and throughout the day by human observers. The results from these measurements are strikingly similar. In the morning, people usually travel alone, which is reflected by the average batch size of 1.1 persons or less, and the high proportion, about 90 percent to 95 percent, of people moving alone. On the other hand, people tend to form large groups of up to five persons during lunch time, which results in the average batch size of 1.4-1.7 persons. Hence, during lunch time, the size of the batches deviates significantly from only one person, which should be taken into account in elevator traffic simulations to obtain reliable results.

Figure 2: The batch-size distribution depends on the time of day. For example, in office buildings, people arrive at work alone in the morning (left) whereas they move in batches during lunch time

Figure 2: The batch-size distribution depends on the time of day. For example, in office buildings, people arrive at work alone in the morning (left) whereas they move in batches during lunch time (right). Source: Kone Corporation

The varying intensity of people flow throughout the day has been studied quite extensively, and the existing elevator traffic simulation models generate the measured daily traffic patterns [12]. However, the batch arrivals seem to be the final piece to match the results of the theoretical models to the real people flow. A good example is the up-peak roundtrip model. To validate the model, its key variables as well as the batch size distribution were measured during the embarkation of a cruise ship [13. During an embarkation, the traffic consists mainly of people traveling from the embarkation decks to the upper decks with some interfloor traffic between the upper decks. This means that the traffic is almost pure up-peak traffic, making the validation possible. Based on the measurements, the average batch size was about 1.7 passengers, indicating that many of the batches were clearly larger than one passenger. A new variant of the classical up-peak roundtrip model, which takes the average batch size as an input parameter, produced results that matched up with the measurements. Since the up-peak roundtrip model can be used to validate simulation models, the new model enables the validation and calibration of simulation models based on batch arrivals.

Figure 3: People flow simulation with KONE BTS.

Figure 3: People flow simulation with KONE BTS. Source: Kone Corporation

Finding the Way in the Virtual Building

The KONE Building Traffic Simulator (KONE BTS) is capable of modeling complex high-rise buildings having multiple elevator groups, staircases and escalators [10]. BTS generates autonomous virtual agents according to the arrival process model and routes them through the building from their origin floors to their destination floors. Each agent belongs to a passenger group that has a set of physical and behavioral characteristics. Examples of physical characteristics are walking speed, space demand and ability to use certain transportation devices. The behavioral characteristics define how the agents weight different routes when making way-finding decisions during a simulation.

BTS generates a way-finding network automatically from building hotspots or nodes. A hotspot is a place where a passenger typically makes a decision about the next destination or hotspot. Example hotspots are building entrances, waiting areas such as elevator lobbies, escalators and staircases. During a simulation, every agent makes a way-finding decision at each hotspot. This decision problem can best be described by multi-attribute value theory, which weights different behavioral criteria or modules according to the passenger group parameters. For example, a module representing the reluctance to walk long distances favors routes that do not contain walking. The final route decision is drawn from a probability distribution where a high route probability corresponds to a high utility. The advantage of this approach is that even if the agents were identical with respect to their parameters, they may choose different routes if they are about equally probable. A deterministic approach would make identical passengers choose the same route, which would become congested while other routes would remain unused.

Figure 4: Snapshots of a trial of the experimental study on pedestrian behavior and exit selection in evacuation of a corridor.

Figure 4: Snapshots of a trial of the experimental study on pedestrian behavior and exit selection in evacuation of a corridor. Source: Kone Corporation

After 9/11, authorities started to pay more attention to evacuation of tall buildings, and elevators became an accepted means for evacuation. In tall buildings, evacuation can be several times faster with the elevators than with two to three staircases as required by the current safety standards. The number and location of evacuation elevators are defined in the building design phase. In the evacuation simulations, passenger waiting times and journey times to the rescue levels, as well as the time to empty the whole building, are considered [11]. When simulating such an evacuation situation, the virtual agents need to decide which of the available transports to use when leaving their home floor. The decisions are based on the behavioral characteristics of the agents and the congestion levels of the transports [15].

Why is the Back Row Rushing?

An important decision evacuees usually face is what exit to use. In agent-based evacuation simulation models, the preferred exit is often selected based on a rule or a more advanced algorithm. The agents may, for example, select the nearest exit or observe the situation and make optimal decisions [2]. It is, however, questionable whether people in a threatening situation would make optimal decisions and select the fastest exit. The information available, for example the length of the queues at the exits, and the time to process it may be limited. Indeed, the results of an experimental study suggest that people in an evacuation situation may not be able to select the fastest exit [5].

Most catastrophes in evacuation situations, like in evacuations from fire, take place just in front of the most used exits. The speed of pedestrian flow through the bottlenecks is one of the factors affecting the outcome of evacuations. The factors affecting these flows have been studied with evacuation experiments and by computer simulations. One of the key findings is the faster-is-slower effect, which says that the harder people push toward an exit, the more people flow through it is reduced. This is due to the increased pressure, which increases the interpersonal friction forces and creates jams and clogging in front of bottlenecks.

Figure 5: FDS+Evac is able to create realistic behavior in a crossing (left). Stable equilibria curves in a spatial game of patient (yellow) and impatient (black) agents under threatening conditions

Figure 5: FDS+Evac is able to create realistic behavior in a crossing (left). Stable equilibria curves in a spatial game of patient (yellow) and impatient (black) agents under threatening conditions (right). Source: Kone Corporation

The widely used social-force model describes the crowd with a self-driven many particle system. It produces realistic flows through bottlenecks and creates the faster-is-slower effect [3]. It does not, however, explain how, why and when the crowd members adopt impatient pushing behavior. The reason for a change in behavior can be in the external conditions or in the behavior of the crowd members. In the literature of social psychology, the pushing behavior is often related to panic. Nevertheless, it has long been understood that actual panic occurs rarely in real crowds and evacuating people tend to behave rationally.

Heliövaara et al. [6] present a spatial game model for pedestrian behavior in situations of exit congestion. The options of the agents are either to act patiently or impatiently, and they play the game against other agents in their surroundings. The payoffs for the agents are derived from natural assumptions on crowd dynamics, which turn out to result in a hawk-dove game, a basic game in evolutionary game theory. Since the parameters of the game depend on the agents’ location in the crowd, the agents further back in the crowd act impatiently and push while the agents in front of the exit play a different game by acting patiently.

The equilibria of the game are studied computationally: The game model is coupled with the social-force model, and this is implemented in a popular fire dynamics simulator FDS+Evac where the agents can rotate their bodies and dodge each other, too [7, 4]. Simulation results show that the model gives a valid explanation for the clogging occurring at bottlenecks of egress routes under threatening conditions.

Marja-Liisa Siikonen (marja.liisa-siikonen@kone.com) is the director of people flow planning at KONE Corporation, headquartered in Helsinki, Finland. Juha-Matti Kuusinen and Janne Sorsa are managers and Henri Hakonen is a senior specialist (people flow planning) at Kone. Harri Ehtamo is a professor at Aalto University.

References

  1. Alexandris, N.A., 1977, “Statistical models in lift systems,” Ph.D. thesis, University of Manchester, Institute of Science and Technology.
  2. Ehtamo, H., Heliövaara, S., Korhonen, T, Hostikka, S., 2010, “Game Theoretic Best-Response Dynamics for Evacuees’ Exit Selection,” Advances in Complex Systems, Vol. 13, pp. 113-134.
  3. Helbing, D., Farkas, I., Vicsek, T., 2000, “Simulating Dynamical Features of Escape Panic,” Nature, Vol. 407, pp. 487-490.
  4. Heliövaara, S., Korhonen, T., Hostikka, S., Ehtamo, H., 2012a, “Counterflow Model for Agent-Based Simulation of Crowd Dynamics,” Building and Environment, Vol. 48, pp. 89-100.
  5. Heliövaara, S., Kuusinen, J.-M., Rinne, T., Korhonen, H., Ehtamo, H., 2012b, “Pedestrian behavior and exit selection in a corridor – An experimental study,” Safety Science, Vol. 50, pp. 221-227.
  6. Heliövaara, S., Ehtamo, H., Helbing, D., Korhonen, T., 2013, “Patient and Impatient Pedestrians in a Spatial Game for Egress Congestion,” Physical Review E, Vol. 87, pp. 012802.
  7. Korhonen, T., Hostikka, S., 2009, “Fire Dynamics Simulator with Evacuation: FDS+Evac - Technical Reference and User’s Guide,” VTT working papers 119, VTT Technical Research Centre of Finland.
  8. Kuusinen, J.-M., Sorsa, J., Siikonen, M.-L., Ehtamo, H., 2012, “A study on the arrival process of lift passengers in a multi-story office building,” Building Services Engineering Research & Technology, Vol. 33, pp. 437-449.
  9. Kuusinen, J.-M., Sorsa, J., Siikonen, M.-L., 2015, “The Elevator Trip Origin-Destination Matrix Estimation Problem,” Transportation Science, Vol. 49, pp. 559-576.
  10. Siikonen, M.-L., 1997, “Planning and Control Models for Elevators in High-Rise Buildings,” Ph.D. thesis, Helsinki University of Technology, Systems Analysis Laboratory.    
  11. Siikonen, M.-L., Hakonen H, 2003, “Efficient Evacuation Methods in Tall Buildings,” Elevator World, Vol. 51, pp. 78-83.
  12. Siikonen, M.-L., 2013, “Traffic Patterns in Hotels and Residential Buildings,” in: Peters, R.D. (ed), Proceedings of 3rd Symposium on Lift and Escalator Technologies, pp. 140-147.
  13. Sorsa, J., Siikonen, M.-L., 2014, “Up-peak Roundtrip Time in Theoretical Calculation, Traffic Simulation, and Reality,” in: Lustig, A (ed), Proceedings of Elevcon 2014, IAEE, pp. 347-357.
  14. Strakosch, G.R., Caporale, R.S., 2010, “The Vertical Transportation Handbook,” John Wiley & Sons, New York.
  15. Susi, T., Sorsa, J., Siikonen, M.-L., 2005, “Passenger Behaviour in Elevator Simulation,” Elevatori, Vol. 34, pp. 28-37.
  16. Tyni, T., Ylinen, J., 2001, “Genetic algorithms in elevator car routing problem,” in: Spector, L, Goodman, E.D., Wu, A., Langdon, W.B., Voigt. H.-M., Gen, M., Sen. S, Dorigo, M., Pezeshk. S.