Title: Discrete/Stochastic Simulation
1Discrete/Stochastic Simulation
2Usages
- Business Process Re-engineering
- Manufacturing Process Design
- Service Process Design
- Operations
- Supply chains
- As a planning tool
- As an innovation and improvement tool
3More usages
- Queuing systems
- More general than analytic queuing theory
- Inventory systems
- PERT networks
4Applications of Discrete Stochastic Simulation
- Resource management systems
- Pollution management systems
- Urban and regional planning
- Transportation systems
- Health systems
- Criminal justice systems
- Industrial systems
- Education systems
- eCommerce systems
5What Discrete Stochastic Simulation isnt
- Stocks, states, rates, flows, information
- Continuously changing variables
- Causal loop diagramming
- stock-and-flow diagramming
6What Discrete Stochastic Simulation is
- Probabilistic occurrences
- Activity completions
- Processes
- Precedence relationships
- Probabilistic routing
- Events, Entities and Attributes
7An exampleSouthwest Airlines aircraft
turn--ACTIVITIES
- Disembark passengers
- Cabin cleanup
- Embark passengers
- Unload baggage
- Load Baggage
- Refuel
- Remove waste
- Refurbish snacks and drinks
8EVENTS for the aircraft gate turn
- Arrival at gate
- Beginning of unloading
- Completion of passenger unloading
- Beginning of cleanup
- Ending of cleanup
- Beginning of passenger loading
- Ending of passenger loading
- Beginning of baggage unloading
- Ending of baggage unloading
9ACTIVITIES EVENTS
- Activities always have time duration.
- That time duration is in general random
- Events are instants in time
- Activities are sometime engaged in by entities
- Activities, events and entities have attributes
10Events, Entities and Attributes
- Entities may be permanent or temporary
- Customers, students, piece parts, messages,
boxes, items,--TEMPORARY - Universities, cities, companies, facilities,
servers, professors, service areas --PERMANENT - Both entities and events possess ATTRIBUTES
- Attributes of a server entitymean service time,
std. dev. of service time, distribution type - attributes of an event--event type, no of
entities assoc. with it
11A typical service system scenario
- In the early morning hours between 7 and 9 a.m.,
arrivals to convenience stores are larger than
normal. If there are more than 6 people waiting
in line, new arrivals will balk and go somewhere
else. People arrive at the rate of 1 every
second, but the time is exponentially
distributed. Patrons shop for a time period that
is uniformly distributed between 3 and 5 minutes.
12Convenience Store Scenario, continued
- It takes the checkout clerk an average of 43 sec
to collect money from a customer and provide them
with a receipt, but this time is normally
distributed with a std dev. of 30 sec. . People
will automatically en-queue themselves in front
of the checkout stand. - The manager can hire a second clerk, who is less
well paid but also much slower
13Convenience Store Scenario, continued
- The store manager is interested in
- the average waiting time of his patrons in the
queue - the average number of customers that balked
14With one store clerk
- average waiting time is 128 seconds
- number of balked customers is 76 out of 1000
customers - Check out clerk is busy 86 of the time checking
out customers
15With two store clerks
- average waiting time is 42 secs for the first
server - 69 secs for the second
- There are no balked customers
- Servers are busy 70 and 40 of the time,
respectively
16How does randomness come into play?
- Probabilistic activity durations
- Probabilistic routing decisions
- Probabilistic arrivals
17How is randomness created within a digital
computer?
- Monte Carlo--the computer generation of random
numbers - Sample the clock?--no--not replicable
- maintain a huge file of random numbers--no
- Takes up too much space in primary memory
- On secondary storage, its too slow
- (when deciding to fetch from disk as opposed to
primary memory, the time required. is 500,000
times longer - Use an ALGORITHM --YES, YES
18Why use an algorithm?
- The sequence it generates will be deterministic
- Doesnt take up much space in primary storage
- Takes up no space on secondary storage
19An Algorithm for Generating Random Numbers
- Must be fast (short and sweet)
- Must be capable of generating numbers that have
all of the characteristics of randomness, but in
fact are deterministic - Multiplicative Congruence is one method
20About Random Numbers
- Uniform on the interval zero to one
- They are completely independent and therefore
un-correlated - We represent them this way U(0,1)
21Multiplicative Congruence Algorithm
- CI1 KCi
- function random(float u, int I)
- I I 1220703125
- if Ilt0 then
- I I 2147483647 1
- else
- U I 0.4656613E-9
- return and end
22Notes
- Generates a uniform sequence on the entire
interval of 32-bit integers--0 to 2147483647 - Maps these onto the real interval of 0 to 1
- If the first multiplication causes integer
overflow, the resultant number I will be
negative--it is made positive by adding the
largest 32-bit integer representable 1 - The last multiplication is like dividing the
number by the largest integer possible - 1/2147483647 .4656613x10 to the minus 9
23You can easily generate random numbers in an
EXCEL spreadsheet using the function RAND()
24What about non-uniform random numbers?
- Exponential
- Normal
- Gamma
- Poisson
- Lognormal
- Rectangular
- Triangular
- Beta
- Poisson
- Binomial
- Hyper-geometric
25ONE ANSWER Use the inverse transformation method
- Every non-uniform random variate has an
associated cumulative distribution function F(x)
whose values are contained within the interval 0
to 1 and whose values are uniformly distributed
over this interval - If x is a non uniform random variate, y F(x) is
uniformly distributed over the interval 0 to 1.
26Strategy
- If the inverse of the cumulative distribution
function F(x) exists so that x F-1(y) can be
determined, then - 1) simply generate a random number uniformly
distributed on the interval 0 to 1 - 2) call this number y and apply the inverse
transformation F-1(y) to obtain a random number x
with the appropriate distribution.
27Exponential Random variates
- 1) generate a random number U using the program
given above - 2)then apply EXPRND -XMEAN ALOG(U)
- (This is simply using the inverse distribution
method)
28When Analytic inverses of the cumulative
distribution function are unavailable
- You can use a table function
- You can use specialized algorithms that have been
developed by academics over thirty-five years of
cumulative research
29Outputs
- The animation reveals bottlenecks, idleness
- We are also interested in
- productivity, cycle time (time in the system),
wait time,blocked time, - number of trips made in a given period of time,
- system throughput within a given period of time
- We can get this from the statistical reports
provided after the simulation is finished
30Discrete stochastic simulation time advance
- Time is advanced from event to event.
- The only time instants looked at are the event
times - The events are stored chronologically in time in
an event file known as an event calendar - Corresponding to each event type is an event
subroutine - When an event occurs, its subroutine is called
31Discrete-stochastic simulation as
- A statistical experiment
- Running times must be long enough to ensure
sufficient samples are collected - Several runs are often averaged together
- The starting random number seeds are changed and
the model is rerun - The basic idea is to get the variance to converge
to the actual real-world variance
32Another scenario
- A mufflers-shocks-brakes shop is turning away
business. It is considering hiring another
mechanic or adding another bay. It currently has
four bays.
33Another scenario
- A shipping company has just picked up additional
customers and needs to add capacity. At its
loading warehouse, it has four loading docks. It
also has 10 trucks. Trucks currently wait upon
return for four hours before they can go out on
another trip. Should the company add docks,
remove trucks, or both.
34Another scenario
- A ski rental shop fits customers for boots and
then skiis. It currently has four people working
in the boot area and four people working in the
ski area. Lines are very long and waiting times
unacceptable. Should the shop hire more help or
just shift some of its existing help from skis to
boots or vice versa.
35GO BACK TO 7-11 STORE example
- Consider a 7-11 store in which the 7-9 a.m.
period is of interest. Management is considering
hiring a second clerk. Patrons arrive at the
rate of one every 45 secs. Arrivals are
Exponentially distributed. Patrons shop for a
period that is uniformly distributed between 3
and 5 minutes. The check out time for each
customer is normal with a mean of 43 secs and a
std. dev. of 30 secs. Customers who encounter
a queue of six customers or more upon arrival
will walk away without shopping.
36What are the activities?
- Arrivals
- Shopping
- Checkout
- What about waiting in Queue?
- This is not an activity
- This is handled automatically by the simulation
37What units on time?
- Secs or mins?
- Lets go with SECONDS
- We must be consistent!!!!
38Must also identify
- Locationspoints assoc with the starting and
stopping events of an activity - Path networkthe network the entity travels
- Resourcespermanent entities that act on ordinary
temporary entities - Processesthe activities
39PROMODEL
- SELECT BACKGROUND--optional
- BUILD--gtlocations
- BUILD--gtentities
- BUILD--gtPATH NETWORK
- BUILD--gtresources
- BUILD--gtprocesses and routing
- BUILD--gtarrivals
- RUN IT
40Locations
- Places where an event of importance to the model
occurs - Like an arrival
- A beginning of customer checkout
- An ending of customer checkout
41Entities
- These are the temporary items that pass through
the model of the system - Widgets, chits
- Mail pieces
- Piece parts
- Students
- Cars
- People
42Path network
- The network that will be followed by the entities
and/or the resources
43Resources
- Mobile permanent entities that can move over a
network
44Processes
- A process is required everywhere the entity
undergoes an operation - An exit process is always required
45Routing
- You must specify how the entities move through
the model - Usually you inform PROMODEL what path network to
use
46Arrivals
- The statistics of the arrival process for each
entity type must be communicated to PROMODEL
47Now lets look at PROmodel
48- Exercise 1. (15 points) A local convenience
store has a self-service island from which it
dispenses gasoline. Two lines of cars may form
on either side of the island. The island will
accommodate no more than two cars being filled
with gas on a single side. There is space for no
more than three cars in each of the two queues of
cars waiting for each of the two service areas.
49 - Cars arrive at the rate of one every minute with
a distribution that is exponential. Service
times are normal with a mean of seven minutes and
a standard deviation of two minutes. Cars will
drive away if more than six cars total are either
waiting or in service (regardless of the line
they are in). Once cars have entered the stores
gasoline facility, they will en-queue themselves
into the shortest queue. Formulate a model in
BLOCKS to determine how many cars are turned away
in a day. For BRANCH/ TRANSFERS, be sure to
indicate the type, such as UNCONDITIONALLY to
block 12. Assuming the store is open 24 hours,
setup the model to determine how many cars are
turned away in one 24-hour day.