Forecasting at Inman Yard

1
Forecasting at Inman Yard
2
Gate Operations

• Trucks checked into the Norfolk Southern system.
• Check-in includes inspection and transaction
  recording
• Intermodal service is scheduled
• Surges of trucks arrive shortly before cut-off
  times.

3
The Problem with Surges

• Traffic problems for the community and legal
  problems for Norfolk Southern.
• Customers vehicles delayed
• Potential loss of customers

4
Objective

• Develop forecasting system that can predict truck
  arrivals and inform terminal staffing to avoid
  long lines and idle workers.

5
Data Available

• SIMS Oracle database
• detailed records of transactions, date and time
  reaching back several months
• aggregate records of daily transactions reaching
  back several years

6
The Data
Major customer leaves
7
Daily Transactions
8
Transactions every 15 min
9
Features of the Data

• Annual seasonality by week of year
• Special seasonality e.g., holidays
• Slight growth trend

10
Issues with the data?
11
A Forecasting Methodology

• Use Exponential smoothing with trend and 52
  weekly seasons to predict weekly volume
• Initialize forecast with 1997-1998 data
• Forecast 1999-2000 data and evaluate errors

12
Weekly forecasts
Average Absolute Error lt 10
Two kinds of forecasts
13
Forecast Day of Week

• Use exponential smoothing with 7 seasons to
  predict the fraction of weekly transactions that
  will occur on a given day of the week.
• Why?

14
Daily Forecasts
Average Absolute Error lt ??
15
Forecasting to 15 minutes

• Suggestions?

16
What we did

• Averaged several weeks of detailed transaction
  data
• Obtained seasonal factors for each 15 minute
  interval (what fraction of the days arrivals
  occurred within that period)
• Allowed the user to correct for the arrival rate
  (vs the service rate)

17
Other ideas?
18
(No Transcript)