Você está na página 1de 7

TECHNICAL ARTICLE

A computer simulation model for the


study of police patrol deployment
Gary M. Kern
Department of Management
College of Business Administration
University of Notre Dame

Introduction

Simulation models have been

suggested as a means of assisting patrol administrators


making deployment decisions.
The simulation model reported
here allows the administrator to
study the performance of several

complex dispatching tactics


including multiple unit dispatching, preemption, and intersector dispatching.
Environmental variables,
such as interarrival and service
times, can be set to reflect any
number of possible probability

distributions. Sample reports


generated by the model are
presented, as is information
regarding the models
verification.

Keywords: discrete-event
simulation, police patrol deployment,

226

dispatching rules, SLAM

The police patrol administrator is faced with the difficult


task of using scarce resources to serve an uncertain demand. The administrator must be concerned with the
performance of his or her scarce resources. However, unlike
private service operations, this performance is often
measured in terms of some value other than dollar profit.
Patrol administrators seek to improve the performance
of their scarce resources (patrol cars and sworn officers) by
improving patrol deployment. Deployment can be divided
into three highly interrelated decisions: patrol sector design,
initial patrol unit geographic allocation, and dispatching.
Numerous management scientists have developed
models to aid the patrol administrator (see, for example, [5],
[14], [8], [9]). The models help administrators improve
patrol deployment. Many are static models. The models are
mathematical programming formulations which attempt to
describe the performance of a given patrol deployment in a
given patrol environment.
Many of these previously-developed static mathematical
models had to rely on simplifying assumptions to make

them tractable. For instance, Chaiken and Dormonts Patrol


Car Allocation Model (PCAM) cannot describe the performance of certain dispatching rules that include intersector

dispatching [3], [4], [6].


Some researchers have designed simulation models of
police patrol operations. Kolesar and Walker created such a
simulation model [11]. Their model did not allow for the
preemption of low-priority calls so that high-priority calls
could be served. Larson developed a discrete-event simulation that was quite sophisticated [12]. It could simulate the
performance of deployments which included both preemption and intersector dispatching.
This paper reports the development of a discrete-event
computer simulation model that extends upon Larsons
work. The current simulation model is able to assign calls

for service (CFS) to more than one patrol unit. Such a


dispatching rule is often employed for dangerous incidents.
The model is also capable of simulating sophisticated
dispatching rules.
The model allows the patrol administrator to define his
or her patrol environment. The administrator then enters a
deployment. The model reports the simulated performance
of the deployment given the environment.
The model is currently designed to operate on a mainframe computer, processing in batch mode. The simulation
language SLAM provides the basis for the model [18]. The
models design also makes liberal use of FORTRAN
subroutines.
The simulation model has been used to develop academic research in the area of patrol deployment [10]. These
studies investigated the relative effectiveness of various
deployments in common patrol settings. Simulated performances of the deployments were collected and statistically compared. Such experiments are helpful for support-

ing patrol administration.


This paper reports key features of the simulation model
and its use. The following sections of the paper will
describe:
~

~
~

the event structure that

modeled,
the general design of the model,
input required of the model,
was

simulated

performance results reported by the model,


assumptions and limitations of the model, and
verification of the model.

Event structure
Police

patrol forces are expected to investigate criminal

incidents, intercede in civil disputes, and prevent the


occurrence of future incidents. Most
requests for service are
handled in a common manner. Figure A displays the typical
process used by patrol forces in response to a call for
service. (Figure A is adapted from a time-frame developed
in Larson, [12].)
This framework was used to identify the important
events for the simulation model. Using the terms that
appear in Figure A, the four basic events modeled by the
simulator are:

1) arrival of a CFS (t5)


2) arrival of an assigned patrol unit at the CFS location (t9)
3) completion of service of a CFS (tid
4) return of patrol unit to patrol area (til).

The model initiates the processing of a CFS with the


arrival of the call. Call arrivals are generated by the model
based upon the users description of demand. When a call
arrives, the model uses the selected dispatching logic to
assign patrol unit(s) to serve the call. The patrol unit(s)
begin travel to the location of the CFS. An assigned patrol
unit then arrives at the scene of the CFS. Service of the CFS
is initiated.
The next event is the completion of service of a CFS.
Upon completion of service, the unit(s) involved must
return to the patrolling area to which they are assigned. The
final event in the process is the arrival of a patrol unit to its
patrol area. The unit is again available to respond to a CFS.

General model

design

designed to follow the common service


described
above. A general flowchart of the
procedure
model is shown in Figure B. After the occurrence of any of
the four major events, a set of decision rules supplied by the
user (researcher or patrol administrator) is activated. The
decision rules govern dispatching and the assessment of
patrol unit availability.
For the purposes of the model, deployment has been

The model

was

divided into five components. Three of the components


describe the dispatching decision rules. This is not to be
construed as in some way &dquo;biasing&dquo; the relative importance
of dispatching to overall deployment. Rather, it allows the
model to recognize that dispatching is a dynamic decision-

making process.
Deployment is expressed as the combination of sector
design, initial allocation, queued call selection, intersector
dispatching, and preemption. Sector design and initial
allocation are static decisions; they are made once per
decision period (be it a scheduled work day, a patrol shift,
or an hour). However,
dispatching is performed throughout
the time period in light of the current patrol unit availability. Thus, the model incorporates the dynamic nature of

dispatching.
The features used to describe a dispatching rule have
been suggested in the previous research on deployment.
The inclusion of factors like intersector dispatching and
preemption have been shown in some cases to improve
performance of patrol operations [10].
The model allows the dispatching rule to incorporate
intersector dispatching. Intersector dispatching permits the
assignment of a patrol unit available in sector A to a CFS
located in sector B. Whether to include intersector assignment in the dispatching rule should be a decision made by
the patrol administrator.

Figure A. Sequence of events for a call for service.

227

Figure B. System flowchart.

Preemption, if included in a dispatching rule, allows the


dispatcher to suspend current service of a low-priority CFS.
Such a suspension of service would make available a patrol
unit to

serve an

urgent call that otherwise would have been

queued.
When a CFS arrives, the dispatcher must decide whether
there are units available to serve the call. There may be
units patrolling at the time; if not, the dispatching rule may
define no patrol unit as available. In such a case, the call is
queued and awaits future assignment. A call selection rule
must be established for the order in which queued calls will
be served.

Input
The model requires a description of the patrol situation it is
to simulate. As mentioned above, the deployment is input
228

in terms of five features. The user must also state a profile of


the CFS demand expected to be encountered. This section
describes the nature of these divisions of the input.
Demand is described by a series of characteristics.
Typically, dispatchers categorize CFS into one of several
incident types. Each incident type is given a code. Incident
type maps to a degree of severity or priority. High-priority
calls (those CFS that are more urgent in nature) receive two
or more patrol units for service. It is assumed that safe
service of the call cannot be completed by a single patrol
unit.
The arrival rate for each incident defines the frequency
with which the incident occurs. The service rate describes
the time required to complete service of a given type
of incident. The distribution of incident location frequency (the sector in which each CFS originates) must also

be described.
Input for the environmental factors discussed above is
expected to include an average measure and the expected
distribution for the factors values. For instance, interarrival
times are usually modeled as exponentially distributed
random variables with a given average value. The simulation model is designed so that a specific subroutine generates a specific environmental factors values. This makes it
very simple to change the model to reflect a change in the
assumed distribution for a given factor. Again, such
flexibility is not always possible in a static descriptive
model.
Sector design is described in terms of average travel time
from one sector to each of the other sectors. A matrix of
these average travel times is developed to express each
sectors relative location in the general geographic region.
Travel time measures the distance from one sectors
centroid to another. The user may assume any means of
measuring this distance (e.g., euclidian distance, rectangular distance).
Travel time is the manner in which sector design affects
the service provided to CFS. Similar to the structure used
for generating demand measures, a travel time distribution
is also the basis for a subroutine which generates random
values.
Initial allocation is expressed as the assignment of patrol
units to the sectors. The user simply inputs the total number
of units and then assigns each one to a specific sector.
Dispatching, as mentioned earlier, is described as a
chosen value for each of three features (queued call selection, intersector dispatching, and preemption). This is not to
imply that intersector dispatching must be included. It

Figure G Input values.

implies that the user must decide whether it should be


included. For instance, intersector dispatching may not be
allowed at all, or it could be allowed for only emergency
calls, or it may be allowed for all calls.
The simulation prepares a report describing the input
values. Please refer to Figure C. This report is a good
summary of the values that the user typically must provide.
The current version of the model assumes that all randomly-generated values are exponentially distributed. But,
as mentioned earlier, it is a simple adjustment to change
this assumption.
Based upon the input, the computer model is able to
simulate the arrival of events, processing events according
to the deployment in effect, and collect performance
measures. The following section will describe the output
reported by the model.
z

Output
The simulation model has been designed to allow for
variable replication lengths and number of replications. The
scheduling of an initial transient state, after which performance collectors are cleared, is also possible. The transient
state length is also a variable.
Although many researchers have studied deployment
and suggested numerous models to support the development of deployment, no one has reported a universallyaccepted means of expressing how well a deployed patrol
force has performed its collective job. No satisfactory means
exists for expressing how many criminal incidents a patrol
force prevents. Expressions of incidents &dquo;solved&dquo; are also
questionable in that the patrol force (and the deployment of
this force) do not have total control over eventual incident
resolution.
For these reasons, alternative measures of overall
performance have often been used. These measures address
the efficiency with which patrol service is delivered.
Efficiency relates to the degree of customer (citizen)
satisfaction with the service rendered [17], [2].
The simulation model reports a collection of these
efficiency measures. They can be divided into response time
and dispatch delay. A complete sample of the output
reported by the model is provided as Figure D(1)-D(3).
These reports were generated by a FORTRAN subroutine
built into the simulation. The format of the reports can be
revised as the user desires by altering this single subroutine.
Dispatch delay, in terms of Figure A, is the amount of
time that elapses between when a CFS arrives and when the
dispatcher assigns the CFS to patrol units (t to t ). It reflects
the time a call remains queued before it is dispatched.
Figure D(1) shows the output displayed about dispatch
delay. The average dispatch delay is reported for each
incident code and sector. A histogram reflecting the number
of calls that suffer long delays is also shown.
Response time (also shown on Figure DO) is the length
of time that a citizen must wait before service of their
incident begins. This would reflect not only the dispatch
delay but also the travel time required for assigned patrol
units to reach the CFS location. Referring to Figure A,
response time is the time from ts to tg. Response time is an
important measure because a deployment may not only
affect dispatch delay but also the patrol unit selected to
serve a given call. The
deployment therefore affects the

229

Figure D(2).

Patrol time by sector and Patrol Unit.

equity is also a deployment performance issue. Figure D(2)


presents the relative amount of time each patrol unit spends
on patrol. Units are arranged by the sector to which they are
allocated, and a patrol time average is reported for each
sector.

The possible inclusion of intersector dispatching into a


dispatching rule presents another performance issue. A
patrol administrator may be concerned with the amount of
time a given patrol unit spends working outside the sector
to which it has been allocated. Figure D(3) displays the
information reported addressing this issue. Each patrol
units time during the shift is shown as percentages of time
spent in each sector. This gives some indication of the
interaction between initial allocation and intersector

dispatching in a given deployment.


Figure D(l). Report of dispatch delay and response time.

Assumptions and limitations

travel time in addition to the dispatch delay. Again,


response time averages and histograms are reported for
each incident code and for each sector.
Dispatch delay and response time measure how quickly
the patrol system reacts to CFS. But the patrol administrator
must also be concerned with other factors. Patrol officers
are expected to patrol in expectation that this activity will
reduce the occurrence of future incidents. Patrol officers are
concerned with the relative amount of patrol time they log
in relation to other officers on duty [1]. Therefore, workload

The nature of the patrol deployment situation required the


forming of some assumptions during the design of the
model. These assumptions do not reduce the viability of the
model as a decision-support tool. However, the user must
recognize these assumptions when interpreting the models
results.
The model assumes that dispatchers are perfect and
make their decisions instantaneously. Human error is not a
factor in dispatcher decision-making. Once again referring
to Figure A, the simulator assumes that no time elapses
from t3 to t6. (The model does not consider times for ti to t3)
These factors would not affect the relative performance of a

230

Verification
The simulation model was verified using two techniques.
First, a special subroutine was constructed. The subroutine,
when called, reports the contents of all files, all variables,
and the event calendar. This subroutine was used in
conjunction with several sets of test data to insure that
event logic executed as expected.
The second verification technique applied the simulator
to the task of modelling an M/M/S queueing system. The
model was provided with input data that described a set of

specific M/M/S systems. Each was run, and performance

collected. Measures of average wait in queue and


average length of queue were then statistically compared
with the analytical values calculated using standard
queueing formulas.
Twenty replications were run at each interarrival rate.
The lengths of the replications were computed using
Fishmans subrun method [7] ; a length of 61000 minutes
was used. A transient period was computed based on the
standard deviation of response time observed in preliminary simulation runs. It was found that this standard
deviation tended to stabilize after 7200 minutes of simulated time. Referring to Table 1, it can be seen that the
model was found to represent these queueing systems quite
well. For nine out of ten interarrival rates studied, the
simulated performance value were not significantly
different from the analytical results for an M/M/S system
(confidence level 98%).
data

was

Conclusion
Figure D(3).

Sector allocation of patrol unit time.

given deployment. Further, the way in which such factors


would contribute to the results reported in the model would
be difficult to quantify.
It is assumed that all patrol units are composed of one
patrol car and one patrol officer. In some cities, two-officer
patrol units have been deployed. Given that one officer is
typically insufficient for the safe service of dangerous
incidents, the model assumes at least two patrol units are
assigned to high-priority calls. To adapt the model for the
assumption that two officers are employed in each unit
would not be difficult.
If preemption is not allowed by the dispatching rule,
certain limitations are imposed on a patrol units availability for service. Patrol units are not considered available
during the travel times both to and from a CFS location.
Although this sounds like quite a strict assumption, it
follows actual experience rather well [1].
Preemption, when allowed, is capable of suspending
service
calls during the delivery of that
service. The model is also capable of suspending service
while the patrol unit is still traveling to the location of a
lesser-priority call.
As mentioned above, these assumptions and limitations
do not pose a threat to the models capability of supporting
the deployment decision process. Rather, the user must
recognize their existence when interpreting the results

This paper has presented a discrete-event computer


simulation that can be used to support the deployment
decision for police patrol operations. The model is very
flexible in terms of the types of deployments that can be
studied. The user also enjoys great flexibility in the patrol
environments that can be defined.
A simulation model can present the effects of system

Table 1. Verification

analysis for simulation.

of lesser-priority

presented.

average measure from 20 simulated replications


*&dquo; the hypothesis (simulation value not significantly different from
analytical) was rejected at a confidence level of 98%

231

complexity that are oftentimes difficult to represent in static


mathematical formulations. Such is the case in police patrol
deployment. The deployment is actually the interactive
result of three decisions: sector design, initial allocation, and

dispatching.
Administrators can apply the simulation to studies of
their particular situation. Environmentally-specific factors,
such as CFS frequency, high-priority call frequency, and
travel time distribution may have a significant impact on
the relative effectiveness of a deployment which includes
preemption or intersector dispatching. Studies like these
may be undertaken through the use of the simulation
model.
The model has proven to be an excellent vehicle for
developing experiments which address the design of
deployment. Through the use of simulation, various
deployments can be routinely tested without actually
affecting patrol resources. The model should prove helpful
in the future to both researchers and patrol administrators.

References
Baker, M. Cops. 1985. Simon and Shuster.
Cahn, M. F. and J. M. Tien. 1981. An Alternative Approach to Police Response:

Wilmington Management of Demand Program.. National Institute of Justice


Report.
Chaiken, J. M. and P. Dormont. 1975. Patrol Car Allocation Model: Executive
Summary. Rand Corporation Report number R-1786/1-HUD/DOJ,
September.
Chaiken, J. M. and P. Dormont. 1975. Patrol Car Allocation Model: Users
Manual. Rand

Corporation Report number R-1786/2-HUD/DOJ,


September.
Chaiken, J. M.; T. Crabill; L. Holliday; D. Jaquette; M. Lawless; and E.
Quade. 1975. Criminal Justice Models: An Overview. Rand Corporation
Report number R-1859-DOJ, October.
Chaiken, J. M. and P. Dormont. 1978. "Patrol Car Allocation Model:
Capabilities and Algorithms." Management Science, August, 1978.
Fishman, G. S. 1978. "Grouping Observations in Digital Simulation."
Management Science, January, 1978.
Green, L. 1984. "Multiple Dispatch Queueing Model of Police Patrol

232

Operations." Management Science, June, 1984.


Green, L. and P. Kolesar. 1984. "Comparison of Multiple Dispatch and M/
M/c Priority Queueing Models of Police Patrol." Management Science,
June, 1984.
Kern, G. M. "The Effect of Intersector Dispatching on Police Patrol
Performance."

Omega, August, 1987.

Walker. 1975. Simulation Model of Police Patrol


Kolesar, P.and W. E. A

Operations: Program Description. Rand Corporation Report number R1625/2-HUD/NYC.


Larson, R C. 1972. Urban Police Patrol Analysis. Lexington Books.
Larson, R C. and E. A. Franck. 1978. "Evaluating Dispatching Consequences
of Automatic Vehicle Location in
Operations Research 5: 11-30.

Emergency Services." Computers and

Larson, R C. 1978. Police Deployment. Lexington Books.


Larson, R C. and M. A. McKnew. 1982. Police Patrol-Initiated Activities
Within a Systems Queueing Model." Management Science, July, 1982.

McKnew, M. 1983. "An Approximation to the Hypercube Model With PatrolInitiated Activities: An
418.

Application to Police." Decision Sciences 14: 408-

Parks, R P. 1976. "Police Response to Victimization: Effects on Citizen


Attitudes and Perceptions." Report number R77-1, Workshop in Political

Theory and Policy Analysis, Indiana University.


Pritsker, A. A. B. 1986. Introduction to Simulation and SLAM II. Halstead Press,
John Wiley and Sons.

Gary M. Kem is an assistant professor of


the College of Business

management for

Administration at the University of Notre


Dame. He received a PhD from Indiana
University in 1985, an MBA from Case
Western Reserve University in 1979, and a BS
from the University of Virginia in 1978. He has
taught at the University of Virginia, Indiana
University, Indiana State University, and the
Shanghai Institute of Mechanical Engineering as well as at Notre
Dame. His teaching and research interests are in the fields of
Operations Management, Management Science, and Management
Information Systems. He is a member of the Institute of Management Sciences, the Decision Sciences Institute, and the Society for
Information Management.

Você também pode gostar