Elevator Scheduling

Ingrid Cheh
Xuxu Liu
05/05/09
Elevator Scheduling Problem
Elevator as a control system
Response time and behavior depends on
programmed algorithm(s)
Different solution depending on building type
and number of elevators working together

In algorithm, assignment of job:

External elevator request
Internal floor request
Our Problem Context
 Setting:
 15 floor school
administration building
 2 elevators

 Goal:
 Analysis of possible
elevator scheduling
algorithms through
simulation
 Find most optimal in given
problem
 Suggest future directions
Tools
Basic graphical user interface & software
Programmed in Pascal in Delphi programming
environment
Strategies, labeled A, B, C & D
User Interface & Software
Strategy A

 Elevators calculate most requested floor

 Each elevator heads to most requested

Reacting to
floor, unless:
modal  Other elevator already heading there
distribution of  Other elevator contains passenger who want to
requests go there

 If no passengers in elevator, elevator will

position itself on floor 8
Strategy B
 If no passenger in elevator, it waits and
goes to the floor where earliest request is
made

 Otherwise, elevator will head to floors

External
requests in
requested in order of passenger entry into
FIFO manner elevator

 If elevator passes a floor where current

passenger has requested to get off,
elevator stops and picks up new
passengers on direction of travel
Strategy C

 Both elevators do the same

 If elevator is empty, it heads towards the

Internal before
earliest external request
external
requests in
FIFO manner  Otherwise, elevator would head towards
desired destination of occupants

 When elevator opens, it picks up

passengers in either direction
Strategy D
 Elevator A
 Initially goes to floor 15 and descends one floor
at a time picking up passengers who are going
down
 Heads straight back up to floor 15 when reaches
floor 1
Round Robin
scheduling
method  Elevator B
 Initially goes to floor 1 and ascends one floor
at a time picking up passengers who are going
up
 Heads straight back down to floor 1 when
reaches floor 15
Performance Metrics

 w: time request
Average Wait
(w)
submission to final
destination arrival
Passengers
Carried
Passengers
Using Stairs
 c: total passengers
(c) (s) carried
 s: total passengers who
Metrics
gave up elevator wait and
used stairs
Passenger’s Power
Satisfaction Efficiency  p: c/(c+s)
(p) (e)
 e: energy efficiency level
Service
Profile
in elevator operation
(chart)
 chart: running profile of
customers served in
different time brackets
Inputs to Simulation
Available Inputs Our Selection of Inputs
 Scenario  Scenario
 Particular pattern of  30 different scenarios
passengers waiting and
choice of floor
 Peak Hours  Peak Hours
 Choice of more or less  Peak chosen
passenger traffic
 Simulation Length  Simulation Length
 Total simulated of each  5 minutes
simulation run
 Boredom Level  Boredom level
 How much time before  Level of 20
switching to stair use
 Power Efficiency
80.00

Results 70.00
60.00
50.00
40.00
73.70

30.00
20.00
41.07 34.77 38.90
10.00

Average Wait 0.00

Strategy A Strategy B Strategy C Strategy D

100.00

80.00

60.00
84.27 75.93 79.93 Passenger Carried
40.00 61.40
40.00
20.00 35.00
30.00
0.00 25.00 34.57
Strategy A Strategy B Strategy C Strategy D 20.00
24.67 29.70 27.60
15.00
10.00
5.00
0.00
Strategy A Strategy B Strategy C Strategy D
Passenger using Stairs
25.00
Passenger's satisfaction
20.00
15.00 19.27 19.73 20.33 80.00
70.00
10.00
12.07 60.00 70.57
5.00 50.00 61.20 57.53
0.00
40.00 57.27
30.00
Strategy A Strategy B Strategy C Strategy D 20.00
10.00
0.00
Strategy A Strategy B Strategy C Strategy D
Analysis of Results
Strategy A is the best strategy
Least average waiting time
Least number of passengers using stairs
Maximum passengers carried
Greatest percentage of passenger’s satisfaction

Ranking of strategies from best to worst:

A, C, D, B
Limitations of Strategies
Strategy A
Efficiency may be related to its position on 8th
floor
Overlooks the potential heavy skew of requests
around the lower floors
Difficult for Elevators A and B both to calculate
the most requested floor
Interlacement of strategies B & C
necessary
“Smart” strategies are less power efficient
Extensions within Problem Context
 Investigation into other variables
Simulation length
Peak hours to non-peak option
Boredom level
Scenario
 Uniform probability distribution is not realistic
 Poisson arrival processes might be more reflected
(iTi, j )n  i Ti, j
p(n)i, j  e
n!
 More requests assigned to ground level
 Different setting
 ± Floors

 ± Elevators
Inspiration from Literature

 Passenger behavior (Susi, Sorsa and Sikonen)

Modeling of diverse traffic flows with passenger
compositions that incorporate physical and behavioral
characteristics
 Complex controllers (Bartz-Beielstein, Preuss
and Markon)
Fujitec
Neural network structure to determine control strategy
 Zoning Policy (Chu, Lin and Lam)
Set of floors divided into blocks
Goal of increasing overall handling capacity
Summary
Insight into elevator scheduling simulation
for a particular setting
Compromise between running efficiency
and power efficiency
Very simple and limited model
Extensions possible, as shown by
literature research
Thank You!