To calculate the performance of a new or refurbished lift installation, we need to estimate likely passenger traffic patterns. These traffic patterns are also recognized by the lifts’ traffic control system which adjusts the dispatcher algorithm accordingly. In this article, the authors summarize current, published knowledge of lift passenger traffic patterns and review these against actual survey results. Having identified a need to improve our knowledge of passenger traffic patterns, various means of collecting this data are discussed, including manual surveys, computer vision techniques, infrared counters and analysis based on data logged by lift traffic control systems or traffic analyzers.
Introduction
Assessment of performance is a crucial element in lift (elevator)
design. If lifts are too small, slow or insufficient in number, passengers have
to wait for excessive periods for a lift to arrive in response to landing calls.
Furthermore, passengers traveling more than a few floors in under-lifted installations
often endure long journey times – the result of the lifts having to stop
to answer other calls at most of the intermediate floors. On the other hand,
the luxury of an over-lifted building is an expensive one. Floor area that could
be let to tenants is lost to additional or larger lift lobbies and shafts, and
capital, maintenance and energy costs of the installation are higher.
The need to specify appropriate numbers of lifts, their capacity and speed,
etc. has led to the study of lift traffic analysis. Lift traffic analysis allows
us to assess the performance of a proposed lift installation based on estimates
of building passenger traffic patterns. Lift traffic analysis techniques, ranging
from up-peak calculations (1,2) to general analysis formulae (3) and simulation
techniques (4), are well developed and widely applied. However, lift performance
results from lift traffic analysis are of no better quality than the estimated
passenger traffic patterns that are used in the calculations or simulations.
In operation, lift control systems adapt to changing demands based
on their designers’ understanding of passenger traffic patterns. Control
strategies appropriate to the current traffic pattern (e.g., up-peak, down-peak
algorithms) can improve performance significantly.
In this article, the authors summarize current, published knowledge of lift
passenger traffic patterns and compare this with survey results. Current design
guidelines are questioned, and means of improving our knowledge of lift passenger
traffic patterns are discussed.
Current Knowledge of Traffic Patterns
General Approach
In estimating prospective passenger traffic patterns, a designer
might consult:
-- Elevator Traffic Analysis Design and Control (1);
-- Vertical Transportation, Elevators and Escalators (2);
-- CIBSE Guide D, Transportation Systems in Buildings (5); and
-- Standards, e.g., in the U.K., BS 5655 Part 6 (6).
There are other sources of information, including manufacturers’
planning guides, but these tend to reiterate the recommendations of above. G.C.
Barney, S.M. dos Santos (1) and G.R. Strakosch (2) present example diagrams
of passenger traffic in a commercial office building. These diagrams have been
re-drawn in Figures 1 and 2.
According to Barney and dos Santos (1), conventional design procedure is to
determine the performance of lift systems for the morning up-peak traffic situation.
This is consistent with the authors’ experience from reviewing consultants’
and manufacturers’ calculations. The common approach is probably because:
-- The up-peak traffic condition is relatively simple to analyze;
-- It is widely accepted that, if a lift system can cope efficiently with
the morning up peak, then it will cope with other periods in the day; and
-- Most traffic flow design recommendations are for up-peak handling capacity
CIBSE Guide D (5) suggests the following up-peak traffic flows for design purposes.
Strakosch (2) places most emphasis on the incoming up-peak traffic but also proposes two-way and outgoing traffic criteria. BS 5655 Part 6 (6) offers only up-peak design criteria.
Published Lift Traffic Surveys
Detailed lift traffic surveys carried out by researchers, consultants and manufacturers are rarely published. One exception is a survey of passenger traffic in two office buildings published by the Building Research Establishment (BRE) in 1974 (7). Results are summarized in Table 2.
The BRE survey also concluded that lunchtime traffic amounts to 12% of building population in both buildings, but this includes stair traffic.
Traffic Surveys
Passenger traffic surveys have been carried out on behalf of the authors at a range of buildings. Results are summarized in Figures 3-7 which record the traffic to and from the main terminal floor(s), except for Building E, where the predominant traffic flow was interfloor. Traffic was measured only during peak periods (normally morning, lunch and evening; morning and evening for the hotel).
Discussion
The traffic survey results suggest that the morning traffic peaks are less marked in buildings than they were when traditional up-peak design criteria were formulated. In work-related buildings occupied during the day, the busiest period appears to be during the lunch period. Lunch traffic is a combination of up- and down-peak traffic to the main terminals, but it often includes an element of interfloor traffic. This interfloor traffic is especially significant in buildings with restaurants, meeting rooms, etc.
It can be shown that if the same total handling capacity is assumed, people wait longer for a lift at lunchtime than they do during a morning up peak. This is because the combination of passengers traveling up and down the building results in more stops per roundtrip. Consequently, the authors suggest that future design criteria for traffic analysis should use the lunch peak as a primary design criterion. It would be dangerous to disregard established up-peak design criteria without a wider study of building traffic flow peaks. Thus, the remainder of this article discusses means of representing and collecting traffic data so that, in due course, updated design criteria can be formulated for a wide range of buildings.
Representing Lift Traffic Flows
Traditionally, lift traffic flows have been defined in terms of
the percentage of the building population transported upward and downward in
five minutes, as used in Figures 1-6. For more complex flows, such as lunch
peaks, we need a more comprehensive way of describing lift traffic. Peters presented
an approach in his paper on general analysis lift calculations (3) that allows
us to describe traffic flow completely. Two terms are required:
-- µi – the passenger arrival rate at floor i (defined for each
floor at which passengers may arrive); and
-- dij – the probability of the destination floor of passengers from
floor i being the jth floor (defined for all possible i and j).
Using these terms, a simple up-peak traffic flow in an office block could be represented as in Figure 8, and a more complex traffic flow could be represented as in Figure 9.
Future design criteria should enable the designer to estimate
peak traffic flows in these terms from a knowledge of the office building population,
number of hotel rooms,
etc. – dependent on the building type. The lift performance can then be
assessed analytically or by simulation.
Carrying Out Lift Surveys
Alternative Survey Techniques
There are a number of alternative approaches to collecting data on lift passenger traffic patterns. Those considered by the authors are discussed in the following subsections. Other and new technologies may yield alternative approaches.
Manual Surveys Using Observers
In manual surveys, observers count passengers in and out of the
lifts. Manual surveys are normally based on one of two approaches:
I. Survey from the main terminal(s), where observers count passengers in and
out of the lifts as they arrive/depart from the main terminal floor(s). Traffic
between other floors is assumed to be negligible. Survey results given in Figures
3-6 were collected using this approach.
II. The in-car survey, where observers are situated in the lift car and count
the passengers in and out at every floor where the lift stops. Survey results
given in Figure 7 were collected using this approach.
Manual surveys are discussed in detail in (1) and (8). The new generation of cheap, miniature video cameras (used with a video recorder) can be used to make observation unobtrusively; the recorded video is played back off-site for counting.
The survey techniques do not allow us to describe traffic flow
completely as:
I. Only measures arrival rate at the main terminal floor(s) and requires assumptions
to be made about arrival rates and destination probabilities on other floors.
These assumptions are generally based on the building floor populations.
II. Measures arrival rates at all floors, providing superior data to (i). Overall
destination probabilities (average over all arrival floors) can be approximated
from the count of passengers as they leave the lift. Collecting data to enable
traffic to be described completely is impractical for the human observer unless
traffic is light. To achieve a full date set of destination probabilities, the
observer would have to track every passenger, e.g., passenger 53 entered the
lift at floor three and alighted at floor six; passenger 54 entered the lift
at floor four and alighted at floor 10, etc.
Control System and Traffic Analyzer Surveys
Conventional Systems
Traffic analyzers are linked to the lift control system and record the time every landing and car call is made and cleared. They analyze this data and provide a range of performance results and graphs. Modern control systems incorporate similar functionality.
A range of traffic and performance measures can be determined,
for example:
-- average response time to landing calls by time of day;
-- distribution of response times; and
-- distribution of car calls by floor.
Traffic analyzers give a good indication of a lift system’s performance but very limited information about the actual passenger traffic flow. This is because they have no means of determining the number of people transported on each trip, e.g., a landing call at floor five and corresponding car call to floor seven could equally be a single person or a group of people traveling together. The use of accurate weighing devices would provide a guide to passenger load. However, ambiguities occur if people are loading and unloading at the same floor, e.g., five people loading and three people unloading would provide the same weight differential as two people loading. Therefore, on its own, traffic analyzer data does not give us the information we require.
Inverse S-P Method
Lutfi Al-Sharif suggested a means of interpreting data that is available to traffic analyzers. The Inverse S-P method (9) applies conventional up-peak traffic analysis formulae “backward” to estimate the number of passengers using a lift from the number of car calls and lift movements. The Inverse S-P method is effective, yet applies only to up- and down-peak traffic.
Estimation of Complete Traffic Flow
Peters reported having derived a method for extrapolating (complete)
traffic flow from control system data in 1994 (10). The development of this
method was halted after successful preliminary tests as further work was impractical
without taking data directly from lift system controllers. Access to the necessary
data was not available at that time. The proposed method is outlined as follows:
-- The passenger arrival rate, µi, is a function of the average time
between a lift leaving floor i traveling up and the up landing call being pressed
by the next passengers arriving at the landing station and the average time
between a lift leaving floor i traveling down and the down landing call being
pressed by the next passengers arriving at the landing station.
-- This function can be derived by applying the assumption that the arrival
of passengers at a lift landing is reasonably modeled by a Poisson process.
(This assumption has previously been applied in lift traffic analysis [1,3].)
-- Destination probabilities can be estimated by analysis of car calls
registered as the lift leaves each landing. Not every passenger will register
a car call (as other passengers will have pressed the button first). However,
over time, the relative frequency of unregistered car calls being pressed will
provide a good indication of the average destination probabilities from each
floor.
Figure 10 records some results from the preliminary tests where control system
data was collected “manually” by observation.
Computer Vision
Researchers (11,12) have applied image-processing techniques to
video pictures of lift lobbies to determine the number of people waiting for
the lifts. This lobby count aids the control system by enabling it to prioritize
calls from busy floors.
A spinoff from the lobby-count system developed at Brunel University was a prototype
“traffic surveyor” to count the passengers as they loaded and alighted
the lifts. The system applies similar image-processing techniques to the lobby-count
system but compares each video frame in sequence to track people across the
scene. If you join or leave the scene from the areas defined as the lift doors,
they are counted as having loaded or alighted the lifts. In tests, the system
was found to be 80-85% accurate, errors being due mainly to a tendency to mis-track
people from one image to the next.
This Brunel University research project has now concluded, so no further development
is envisaged. Image processing is an active research area, and improved pedestrian
tracking systems are likely to be developed in the future, probably initially
for security applications. In due course, we are likely to be able to purchase
general-purpose pedestrian tracking systems that will provide us with the basis
for complete measurements of traffic flow.
Infrared
Infrared technology is widely applied, particularly in the security industry. Traffic surveys using photocells or infrared beams were suggested in (13,14). The approach requires a minimum of two horizontal beams to count people passing through the detection field in single file. The sequence of beam states enables direction to be determined. If people are walking side by side, horizontal beams will detect only a single person. This can be overcome by mounting beams vertically. A system believed to be using this approach is installed in a London department store monitoring escalator traffic.
Initial lab and site tests suggest that, although system logic can be fooled, in practice, the overall counting accuracy of infrared counting systems is high. The infrared detectors effectively replace observers in manual surveys, so the data collected does not describe traffic flow completely. However, infrared technology is available and relatively inexpensive to implement.
Written Surveys
Written surveys, where people record the times of lift trips on a form, have been found to be unreliable (7). This was confirmed from the results of a written survey at Building A (Figure 3). This is probably due to a tendency for people to record their arrival and departure times as the fixed working hours of a company.
Security Systems
Various security systems are applied to control access in buildings, some of which are integrated with the lift systems. Systems that use swipe cards to call the lift, or a key pad to control access to specific floors, do not yield useful traffic flow data. Where they are installed, systems that identify passengers individually as they arrive and depart lift lobbies will enable traffic flow to be monitored completely.
Other Issues
Use of Stairs
In planning lift installations, some designers make allowance
for the use of stairs. The authors’ survey experience suggests:
-- The number of people using the stairs in lieu of the lifts drops sharply
as the journey travel increases.
-- People are less likely to walk up than down.
-- An attractive staircase sited adjacent to the lifts is far more likely
to be used than a back staircase.
In the Building C (Figure 5) survey, use of the staircase was virtually nil in spite of the lifts being heavily loaded and long passenger waiting times. The main staircase was an unattractive fire escape sited well away from the lift lobby. Figure 11 shows the associated stair usage for the BRE (7) and Building A (Figure 3) surveys.
In lift traffic surveys, we need to assess stair usage. Otherwise,
generalized recommendations will be inappropriate to:
-- High-rise buildings where the relative use of stairs is far less significant;
and
-- Buildings where stair access is poor.
Occupancy
If the results of traffic surveys are to be applied in the design of other buildings, it is important that traffic is recorded relative to the actual building population. Plotting survey results of a partly occupied building relative to nominal building population can suggest misleadingly low traffic flows.
Conclusions
It is important for lift designers to have a good understanding of lift traffic flows. Most lift installations are designed on the basis that the morning up peak is the most onerous traffic condition for lifts; yet, traffic surveys suggest that the lunch period is more often the busiest.
In planning new lift installations, it would be dangerous to disregard conventional up-peak design criteria completely without a wide study of other traffic flow peaks. However, in many cases, designs applying up-peak traffic analysis are inappropriate.
A range of surveying techniques has been reviewed as a means of establishing passenger traffic flows. The authors continue to collect data and encourage others to publish their results so that improved design criteria can be established. Survey techniques are improving. Currently, the authors favor the infrared system as the best available technology.
Improved knowledge of traffic flows will also aid control system design.
Acknowledgements
The authors would like to thank the Engineering and Physical Sciences Research Council, The Ove Arup Partnership and CIBSE for their financial support of this research.
References
1. Barney, G.C. and S.M. dos Santos. Elevator Traffic Analysis
Design and Control, 2nd Edition. (London: Peter Peregrinus) (1985).
2. Strakosch, G.R. Vertical Transportation: Elevators and Escalators, 2nd Edition.
(New York: J. Wiley & Sons Inc.) (1983).
3. Peters, R.D. “The Theory and Practice of General Analysis Lift Calculations.”
Elevator Technology 4, Proceedings of ELEVCON ‘92. (The International Association
of Elevator Engineers) (1992).
4. Jenkins, K. “Elevator Simulation Techniques.” Elevator Technology
4, Proceedings of ELEVCON ‘92. (The International Association of Elevator
Engineers) (1992).
5. Various Authors. CIBSE Guide D, Transportation Systems in Buildings. (The
Chartered Institution of Building Services Engineers) (1993) ISBN 0 900953 57
8.
6. “BS 5655 Part 6: Lifts and Service Lifts: Part 6, Code of Practice for
Selection and Installation.” (London: British Standards Institution) (1990).
7. Courtney, R.G. and P.J. Davidson. A Survey of Passenger Traffic in Two Office
Buildings. (Watford: Building Research Establishment) (June 1994).
8. Various Authors. ELEVATOR WORLD’s Guide to Elevatoring (Elevator World,
Inc.) (1992).
9. Al-Sharif, L. “New Concepts in Lift Traffic Analysis: The Inverse S-P
Method.” Elevator Technology 4, Proceedings of ELEVCON ‘92 (The International
Association of Elevator Engineers) (1992).
10. Peters, R.D. “Green Lifts?” Proceedings of CIBSE National Conference
1994. (The Chartered Institution of Building Services Engineers) (1994).
11. So, A.T.P. and S.K. Kuok. “A Computer Vision Based Group Supervisory
Control System.” Elevator Technology 4, Proceedings of ELEVCON ‘92.
(The International Association of Elevator Engineers) (1992).
12. Schofield, A.J., T.J. Stonham and P.A. Mehta. “A Machine Vision System
for Counting People.” Proceedings of Intelligent Buildings Congress ‘95
(Israel: The Stier Group Ltd.) (1995).
13. Kaakinen, M. and N.R. Roschier. “Integrated Elevator Planning System.”
(ELEVATOR WORLD) (March 1991).
14. Siikonen, M.L. “Simulation – A Tool for Enhanced Elevator Bank
Design.” (ELEVATOR WORLD) (April 1991).