UMR Statistical Analysis

1
UMR Statistical Analysis

• Inland Navigation Appointment System Study
• Upper Mississippi River Locks 20-25
• Center For Transportation Studies
• University Of Missouri, St. Louis
• 15 June 2005

2
Phases of UMR Statistical Analysis

• Produce performancebenchmarks with OMNI data
  under historical operating rules and physical
  conditions
• Create system for development and maintenance of
  sets of statistical models to support enhanced
  simulation of UMR traffic flows and lockage
  operations
• Provide comparisons of simulated system
  performance under alternative operating
  procedures against historical bench marks

3
Performance Bench Marks

• System status through the shipping season
• Monthly time-weighted averages and monthly
  maxima for
• Number of vessels queued for lockage at each lock
  (upbound and downbound)
• Status of locks (percentage of time busy versus
  idle)
• Number of vessels under way in each river pool
  between locks (upbound and downbound)

4
Performance Bench Marks (Continued)

• Lock operations
• Time spent waiting by vessels from arrival at
  lock to beginning of lockage
• Time spent by vessels in lockage operation
• Throughputs of vessels and tows
• Vessel transit time
• From completion of current lockage to arrival for
  next lockage
• Vessel itineraries for selected periods
• Generated for chosen vessels for comparison
  against actual vessel tracking data - validation
  of OMNI data

5
General Observations From Descriptive Statistics
for 2003

• Based on 104,730 records of non-recreational
  lockage records for the UMR system
• Average utilization (percentage of times locks
  were engaged serving vessels) peaked in summer
  months at values between 74 and 79 for locks 20
  through 25
• Maximum queues for lockage ranged between six and
  eight vessels with tows
• Vessels spent 1.5 of their time (from first
  recorded UMR lockage to last recorded UMR lockage
  in the shipping season) waiting for lockage
  somewhere in the UMR system and 1.7 of their
  time undergoing lockage somewhere in the UMR
  system

6
General Observations (Continued)

• While there were considerable delays in locking
  vessels in the congested areas of the river, the
  delays constitute a small percentage of the
  annual vessel operation times
• Limited potential of increasing the utilization
  of towboat resources by
• using alternative sequencing rules
• or increasing lock capacity at the five
  bottleneck locks
• unless volumes of river traffic increase
  substantially above 2003 levels

7
Analytical Issues

• Seasonal variation in numbers of tows,
  characteristics of tows and efficiency of lock
  operations
• Extent to which OMNI data accurately reflect
  vessel movements and lockage operations
• Interpretation of start lockage times at various
  locks
• Possible variation in the reporting points for
  vessel arrivals at locks (with alleged tendency
  to call in at an earlier point when there are
  several vessels ahead)
• Occasional data entry errors or omissions
• Single vessel numbers recorded for lockages of
  several vessels

8
Analytical Issues (Continued)

• Extent to which vessels currently reduce speed
  to save fuel when delays at locks are
  anticipated
• Extent to which delays are caused by impairments
  at locks (due to weather or other causes)
• Extent to which transit times are affected by
  other river conditions

9
Alternative Structures For Enhanced Simulation
Model

• Initial version (December, 2004)
• All tows except recreational vessels cycle
  through the entire section of the river between
  Lock 25 and Lock 20.
• Recreational vessels are generated randomly for
  lockages according to seasonal patterns

10
Alternative Structures For Enhanced Simulation
Model (Continued)

• Enhancement Level 1 (June 2005)
• Vessels enter the system northbound at Lock 25
  randomly according to seasonal patterns and
  southbound at Lock 20 randomly according to
  seasonal patterns. Alternative configurations
  are (1) double, (2) single, (3) single with
  jackknife lockage, (4) single with knockout
  lockage, or (5) other, including recreational
• At each intermediate lock, a vessel may either
  (1) continue to the next lock in the same
  configuration, or (2) be removed from the system
  and handled through random generation of other
  vessels that appear for lockage according to
  seasonal patterns

11
Alternative Structures For Enhanced Simulation
Model (Continued)

• Enhancement Level 2 (future implementation)
• Vessels enter the system northbound at Lock 25
  randomly according to seasonal patterns and
  southbound at Lock 20 randomly according to
  seasonal patterns
• At each intermediate lock, a vessel may either
  (1) continue to the next lock in the same
  configuration, (2) stop for possible change in
  configuration and proceed in the same direction
  to the next lock, (3) stop for possible change in
  configuration and return to the same lock for
  lockage in the opposite direction, or (4) be
  removed from the system and handled through
  random generation of other vessels that appear
  for lockage according to seasonal patterns

12
Statistical Models To Support The Enhanced
Simulation Models

• Logistic models for likelihoods of alternative
  dispositions of each vessel (transition
  probabilities) after completion of lockage
• Give likelihoods of transition to alternative
  configurations and locations of next lockage
• Regression models for average time required to
  complete the lockage of a vessel
• Regression models for average transit times (from
  completion of current lockage to arrival for next
  lockage)
• Total transit times in pools, including stop
  times for vessels that stop en route from current
  lockage to location of next lockage

13
Factors Considered in Statistical Models

• In logistical models for determining next lockage
  location and tow configuration
• Month of year
• Current tow configuration
• In regression models for average lockage time
• Month of year
• Tow configuration (double, single, jackknife or
  knockout)
• Proportion of lockage that occurs at night
  (suppressed in current version of simulation
  model)
• Whether exchange or turn-back occurs

14
Factors Considered in Statistical Models
(Continued)

• In regression models for average transit time
• Month of year
• Changes in tow configuration and location of next
  lockage
• Percent of journey occurring at night (suppressed
  for current version of simulation model)
• Percent of journey during which impairment is
  experienced at the next lock (suppressed for
  current version of the simulation model)

15
System For Generating and Updating Statistical
Models

• Written as SAS (Statistical Analysis System)
  macros that generate model equations and
  automatically write them to ASCII files for
  importation into the simulation model

16
System For Generating and Updating Statistical
Models

• Macro parameters allow user to determine
• Lists of possible explanatory variables to be
  used in constructing models
• Levels of statistical significance required for
  individual explanatory variables
• Beginning and ending dates of data used to
  calibrate the models
• continued

17
System For Generating and Updating Statistical
Models (Continued)

• Percentiles to be used for screening out unusual
  (extreme) observations when constructing
  regression models (e.g., Excluding observations
  below the 1st percentile or above the 99th
  percentile)
• Whether to include recreational lockages in
  developing equations for lockage times
• Whether to produce corresponding random arrival
  rates (and average inter-arrival times) and
  distributions of configuration types at monthly
  or weekly intervals for each direction at each
  lock

18
Incorporation of Information re Ambient River
Conditions and Lock Impairments

• Seasonal distributions of times between
  breakdowns at locks and duration of breakdowns at
  locks
• Incorporated into the simulation model as
  independent events at each lock according to
  seasonal data
• Consideration of sunrise and sunset times in
  determining the percentage of time that an
  activity occurs in daylight (or conversely, at
  night)
• Allowed in some statistical models but yet not
  incorporated into the simulation model
• continued

19
Incorporation of Information re Ambient River
Conditions and Lock Impairments (Continued)

• Consideration of the percentage of vessel
  transit time that occurred during an impairment
  at the destination lock
• Allowed in some statistical models
• Not incorporated into the simulation model
• Water level data and flow rates did not
  materially enhance the statistical models and
  were not complete for all pools