Transcript Revenue Management in Indian Railways

A Passenger Revenue Management System
(RMS)
for a National Railway in an Emerging Asian
Economy (NREAE)
Goutam Dutta1
Priyanko Ghosh1
1Indian
Institute of Management
Ahmedabad-380015
India
Outline
•
•
•
•
•
•
•
•
Introduction and Motivation
Literature Search
Current Reservation System in NREAE
Optimization Model
Simulation of Passenger Demand
Forecasting Module
Expected Marginal Seat Revenue Approach
Recommendations
Full paper on this research is available from this journal.
A passenger revenue management system (RMS) for a National
Railway in an Emerging Asian Economy
Goutam Dutta, Priyanko Ghosh
Journal of Revenue and Pricing Management 11, 487-499 (6 April 2012)
doi:10.1057/rpm.2012.10 Research
Researchers not able to get a copy of this paper may directly contact the
first author at [email protected]
Introduction and Motivation
•
•
•
•
•
Revenue Management System
Work done by SNCF
Several steps taken by NREAE
A meeting with board member
Visits by Officials of NREAE
Introduction to NREAE
1. One of the five largest in the world
2. Runs 14000 trains daily with 9000 passenger
trains
3. 30 million passengers travel daily in 7083
stations
4. Revenue about 19 billion USD
Challenges of NREAE
• The passenger segment is facing challenges
from low fare airlines which promise
customer satisfaction and less travel time
• The freight sector is facing challenges from
trucks and other road carriers
Industries where Revenue Management applies
•
•
•
•
•
•
•
Perishable Services (Products)
Identification of Market Segmentation possible
Demand is Uncertain and Fluctuating
Fixed Capacity
High Fixed Cost
Low Marginal (Variable) Cost
Advanced Reservation Possible
7
Revenue Management System
Current Data
Historical Data
Forecaster
Optimizer
System
Recommendations
Performance
Monitoring and
Reports
Revenue Management in Railways
•
As per Talluri and Van Ryzin, (2004) following
railways are using RM
•
AMTRACK
•
SNCF
•
Eurostar
•
VIA Rail Canada
Literature Search
Literature Search
• There are about 25 papers on applications of
OR/MS in railroad
• These papers deal various topic related to
IE/OR/MS and not in revenue management
• A few applications in railways revenue
management
Literature Search
• Williamson (1992) formulates two mathematical programming models
for network revenue management (stochastic and deterministic) but
finds no significant difference between them
• Ciancimino et al. (1999) formulate a deterministic linear programming
model and a probabilistic non-linear programming model for railways
and show that the probabilistic model generates more revenue than the
deterministic model
• Boyar (1999) analyzes the seat reservation problem by considering two
scenarios – the price is the same for all tickets and it is proportional to
the distance - and solves the problem by considering deterministic and
concrete algorithms
• Bharill et al. (2008) apply revenue management principles on one of the
trains of Indian Railways. They suggest a differential pricing strategy
on the basis of passenger demand estimates to increase railway revenue
Comparison of Implementations
Objective
Impact
SNCF
To face the competition from
European railways and airlines they
developed well designed decision
support systems to attract
passengers
- Revenue increased by 10 million
francs
- Costs decreased by 3%
Canadian Railways
To create a scheduling system for
railways
- Cost reduced by $300 million
- Total savings exceeded half a
billion dollars
Netherlands Railways
To develop a new timetable that
helps to operate a higher number of
trains and causes fewer train delays
Passenger traffic increased by 15%
- 87% of the trains reach their
destinations within 3 minutes of
their scheduled time
- Annual profit is around E10
million in 2007
German Railways
Increase revenues and reduce costs
through better capacity utilization
- Around 1.5% revenue increased
- Relief trains are 30% lower in
2003 than in 2002
- Standing minutes are reduced by
16% in 2003 than in 2002
Revenue Management System
Current Data
Historical Data
Forecaster
Optimizer
System
Recommendations
Performance
Monitoring and
Reports
Various Fare Classes of NREAE
Codes
Extensions
1A
First class air-conditioned
2A
Air-conditioned 2-tier sleeper
FC
First class
3A
Air-conditioned 3-tier sleeper
CC
Air-conditioned chair-car - only sitting accommodation
(individual chairs) is provided
EC
Executive class, or First class air-conditioned chair-car only sitting accommodation (individual chairs) is
provided
SL
Sleeper class
2S
Second class sitting - only sitting accommodation with
bench style seats
Current Reservation System
• NREAE has two types of reservation systems –
1. PRS (Passenger Reservation System) and
2. UTS (Unreserved Ticketing System)
• 85% of tickets are booked by the PRS and 15% of tickets
are booked by the UTS
• The revenue of NREAE is approximately Rs 931.59 billion
(19.13 billion USD) from which one third is earned from
passenger coaches and two thirds from freight
• The Passenger Reservation System (PRS) offers reserved
seats to passengers in any train from any counter of the
country
Current Reservation System
• Advance booking starts 60 days prior to the day of
departure for all fare classes and for all trains
• The advance reservations are made on FCFS (First Come
First Serve) basis
• NREAE has introduced a booking system called Tatkal
(urgency based scheme) where one can book tickets two
days in advance of the day of departure by paying an extra
charge
• A passenger who books ticket in Tatkal, has to pay the
total fare from origin to destination and as Tatkal quotas
are usually filled up Tatkal earning is constant
Railways booking centers offer seats to
Passengers in FCFS basis
Booking Process
Availability of seat
Non availability of seat
Confirmed tickets are issued on a regular
basis
Passenger asks for Tatkal (Urgent) quota
(2 days prior of the journey date)
Availability of seat
Non availability of seat
Confirmed tickets are issued on Tatkal
Tickets are not confirmed but overbooked in
Reservation Against Cancellations (RAC) and
Waiting List (WL) format
RAC ticket holders can board a reserved coach but are only assured sitting
accommodation even if there are no cancellations.
WL ticket holders are not even guaranteed such sitting accommodations and are
entirely unconfirmed at the time of booking.
Cancellations occur and RAC and WL ticket holders get converted to confirmed
tickets subject to the order of booking.
Revenue Management System
Current Data
Historical Data
Forecaster
Optimizer
System
Recommendations
Performance
Monitoring and
Reports
Optimization Model
INDICES
•
•
•
•
i: Origin indexed by i
j: Destination indexed by j
k: Fare class indexed by k
t: Time period indexed by t
SETS
• S: Set of all stations (1,2,3, ……..n)
• L: Links {(i,j) i  S, j  S, i<j} for all the origin destination
pair
• K: Set of fare classes (k=1,2,3,…….p)
• T: Set of time period (t=1,2,3,…….q)
PARAMETERS
Rt =
ijk
EC t =
ijk
ED t =
ijk
Revenue for fare class k  K for leg (i,j)  L and for time period t  T
Expected Cancellations for fare class k  K for leg (i,j)  L and for time period t  T
Expected Demand forecasted for fare class k  K for leg (i,j)  L and for time period t 
T
CT t = Total Capacity of the Train for time period t  T
C t = Non Tatkal booking allowed for fare class k  K and for time period t  T
k
T t = Tatkal booking allowed for fare class k  K and for time period t  T
k
Ca = Cancellation charges for fare class k  K
k
VARIABLES
X t = Number of tickets to be allocated for fare class k  K and leg (i, j)  L and for time period t  T
ijk
Y t = Boolean variable for fare class k  K and seat number l and leg (i,j)  L for time period t  T
ijkl
Y t =1 if a seat number l is utilized for fare class k  K , leg (i,j)  L and for time period t  T
ijkl
= 0 otherwise
Objective Function
j 1 n
p
q
t t
t
    [ R X  Ca EC ]
k ijk
i  1 j  2 k  1t  1 ijk ijk
+
p
q
t t
k  1t  1 k 1nk

 T R
• First part is the revenue earning from passenger allocations
• Second part is the revenue earned from cancellations that is
a constant term
• Third part is the revenue earnings from Tatkal that is a
constant term
Subject to :
Total Capacity Constraint
Non Tatkal booking is less than the difference between Total capacity and Maximum
Tatkal booking allowed
p
 Ckt
k 1
p
t
<= CT t -  Tk for all k  K and all t  T
k 1
Demand Constraint
Allocated seats should not exceed Expected Passenger Demand
Xt
ijk
t
<= EDijk
for all i,j  L , k  K and for all t  T
Capacity Constraint:
w n
t
  Xijk
i1 jw1
Ct
k
<=
for all i,j  L, k  K, (w=1,2….n-1) , and
for all t  T
Stations:
●
●
1
2
●
3
●
4
● ………….●
5
n-1
●
n
Xij = passenger boarding from source station i to destination station j.
Capacity Configuration:
Station 1:
X12 + X13 + X14 + X15……+X1(n-1)+ X1n <= Ck
Station 2:
X23 + X24 +X25 +......+ X2(n-1) + X2n+ X13 + X14 + X15 +…….+X1(n-1) +X1n <= Ck
Station 3:
X34 + X35 +......+ X3(n-1) + X3n + X14 + X15 +……+.X1(n-1) +.X1n + X24 + X25 +......+ X2(n-1) + X2n <= Ck
……..
……..
Station n:
X1n + X2n + X3n + ……+X(n-1)n <= Ck
A seat can be utilized maximum 7 times
Yt
– Boolean variable =1 if a seat number l is utilized for leg (i,j)  L ,fare class k  K,
ijkl
and for time period t  T
=0 otherwise
w n
t
  Yijkl
<= 7
i1 jw1
k t
 Yijkl =
l 1
Xt
ijk
for all i,j  L , k  K ,(w=1,2….n-1) , and t  T
for all i,j  L ,k  K and for all t  T
Non Negativity Constraint:
Xt
ijk >= 0
for all i,j  L , k  K and for all t  T
Optimization Model
• We collect data of train no.2901 which runs from a metro to
a mini metro over the year 2008
• We consider maximum passenger allocations for an origin
destination pair as forecasted demand data
• We solve the model in AMPL
(A Mathematical Programming Language) and CPLEX solver
version 11.2
• The model was run for four fare classes, 14 stations and one
day, in AMPL/CPLEX 11.2
• The adjusted problem deals with 54694 variables (54528
binary and 166 linear) and 15828 linear constraints (462750
non-zeros) and 1 linear objective (116 non-zeros)
Optimization Model
• The average optimal daily revenue comes to around
Rs 509272
• We consider it as base stage and increase the
passenger demand by 10% in five stages
• In each stage revenue is increased
• Optimal revenue depends on passenger demand
• So accuracy of forecasting of passenger demand
plays a crucial role in optimization model
Revenue Management System
Current Data
Historical Data
Forecaster
Optimizer
System
Recommendations
Performance
Monitoring and
Reports
Simulation of Passenger Demand
• Uncertainty is a crucial feature of passenger demand
• We conduct a simulation study to capture this stochastic
nature of demand
• Passenger demand follows normal distribution (p-value of
KS statistic is <0.01)
• For one year period we compute the mean and standard
deviation of passenger demand of origin destination and
use as inputs in simulation
• We simulate passenger demand for 100 times for each fare
class and for origin destination and build the demand
matrix
Simulation of Passenger Demand
• We refer this as stage 0 or the base stage
• We run our optimization model with these
demand matrices for 100 times and compute
the optimal revenue
• We increase the mean and standard
deviation by 10% in each stage and simulate
passenger arrivals for each fare class and for
origin destination.
Simulation of Passenger Demand
• Stage 1: Simulated passenger demand with 10% rise in mean
and standard deviation of passenger demand of Stage 0
• Stage 2: Simulated passenger demand with 10% rise in mean
and standard deviation of passenger demand of Stage 1
• Stage 3: Simulated passenger demand with 10% rise in mean
and standard deviation of passenger demand of Stage 2
• Stage 4: Simulated passenger demand with 10% rise in mean
and standard deviation of passenger demand of Stage 3
• Stage 5: Simulated passenger demand with 10% rise in mean
and standard deviation of passenger demand of Stage 4
Revenue Management System
Current Data
Historical Data
Forecaster
Optimizer
System
Recommendations
Performance
Monitoring and
Reports
Forecasting Module
• We choose train no.2901 running between a
metro and a mini metro
• We use April 2005-07 booking data as
inputs to predict the passenger arrivals of
April 2008
• As maximum passengers travel from origin
to destination we concentrate on that sector
and do our analysis
Forecasting Module
• We collect data from CRIS (Center for
Railways Information System) on passenger
arrivals for each fare class and for all origin
destination pairs
• The key elements of the data format
includes journey date, booking date, class,
passenger source, passenger destination and
booked passengers
Forecasting Module
• From this format we generate two major
variables for each fare class
(1) Days before departure
(2) Cumulative booking of passengers
• We build booking curves for each fare class
and for origin and destination for April
2005-08
Forecasting Method
• Analyzing the booking curves we divide the booking
horizon into six parts
• D-21(21 days prior to departure), D-14(14 days prior to
departure), D-7 (7days prior to departure), D-2(2 days
prior to departure), D-1(1 day prior to departure) and
D0(day of departure)
• We use additive and incremental pick up methods to
forecast final day bookings of April 2008 for each
fare class
• We measure the forecast accuracy by Mean Absolute
Deviation (MAD) and Mean Absolute Percentage
Error (MAPE)
Error Measurement
• Mean Absolute Deviation (MAD)
• Find absolute difference between forecast and actual
• Average over all observations
• Mean Absolute Percentage Error
(MAPE)
• Find absolute difference between forecast and actual
• Find percentage of actual
• Average over all observations
Forecasting Module
• These forecasting methods work efficiently
in case of 2nd AC and 3rd AC followed by
sleeper but not accurate for 1st AC
• Incremental performs better than additive
method
• It produces MAPE less than 10% for 1,2 or
7 days prior to departure
Forecasting Module
• Mean point forecast is difficult to predict
• We calculate the forecast ranges of
passenger arrivals based on the standard
deviation of historical passenger bookings
and check the percentage of forecast
accuracy
Recommendations
• Forecasting module performs well for 1 or 2 days
prior to departure
• So intermediate stations quota can be released 1 or 2
days before departure date
• Excess demand of a train can be shifted to another
train sharing the same origin and destination
• If no show information is stored in the data
warehouse we will get better patterns regarding
passenger behaviour and can analyze booking
process more efficiently
Thank You