You are on page 1of 21

Determining the Accuracy of

Vessels’ Estimated Time of Arrival


(ETA) with Different Output
Parameters

Akintola Aremu,

Texas Southern University


Outline
 Introduction

 Problem statement

 Estimated Time of Arrival (ETA)

 Previous studies

 Improving performance with data

 Data collection & preprocessing

 Methodology

 Results
 Observation
Introduction
 Multimodal Shipping

• multiple modes
• single document

 Synchronization

• Effective communication
Source: Mediteran Multimodal
• Use of technology
• Data analysis
Problem statement

 Poor alignment in Planning for arrival and departure of


different modes which results in wait

• Deviation of transportation plan


• Truck port congestion
• Reliability of shipping line
• Revenue loss
• Demurrage
Estimated Time of Arrival (ETA)

 Historical pattern
 Statistical Analysis
• Regression
 Machine learning

• Supervised
1. Decision trees 4. Naive Bayes
2. Support vector machine (SVM) 5. Random forest
3. Artificial Neural Network (ANN) 6. Logistic regression
Previous studies
Author Subject Method Result

Parolas I. The effect of Neural network Weather condition has no


(2017) weather condition and support impact on predicted ETA
on ETA vector machine
Fancello G. et Examine calibration Neural network Reduction in the uncertainty
al (2011). of neural network interval of ships’ arrival time
based simulation in port
model for ETA
Pani C. et Categorizing the Naive Bayes, Random Forest algorithm
al.(2014) delay level Decision Trees and performed best in
Random Forests determining the delay level
category
Pani C. et Predict the delay Logistic Regression Qualitative estimate of
al.(2015) and early arrival of Classification Tree whether or not an incoming
vessel Random Forest vessel will likely arrive before
or after the scheduled ETA
Veldhuis,(2015) Automating existing Web Scrapping A prototype that shows the
ETA determining possibilities for a full
method automation of the process
was developed
Improving performance with data

 Algorithm tuning

 Spot check algorithm

Data

 Data filter and Normalization

 Output parameter change


Data collection (PortVision) Data
collection

Preparation
preprocessing

• Fleet Management
GPS
Normalization MLP neural
Zone creation
S-AIS
network
• Control Centers

1. Galveston Sea Buoy Important


attributes

2. Barbours Cut Central


Server
MLP neural
3. Bayport network

Alert Creation
AIS coast Station Ground Station
Accuracy
• Accuracy
calculation
Out of AIS Coast Range AIS Coast Station Range calculation

GPS STANDARD DATA FEED TO


SATELITE
MASSAGE AIS CONTROL
MASSAGE
MASSAGE CENTER

Compare
 Data from last port of call to pilot point result

 Data from pilot point to Terminal


Input Parameter Data
collection

Variable type To Pilot point To Terminal


preprocessing

IMO IMO
MLP neural
Identification and culture Call Sign Call Sign
Normalization
network

Vessel Owner Vessel Owner Important


attributes

Length Length
MLP neural
Width Width network
Physical structure
Capacity [TEU] Capacity [TEU] Accuracy
Accuracy
calculation
calculation

Data collection time Data collection time


Compare
result
ETA Berth ETA
Time
Actual time of Arrival Actual Berth time
(ATA)
Input Parameter Data
collection

Variable type To Pilot point To Terminal preprocessing

Latitude Terminal
Longitude
Vessel position Normalization MLP neural
network
Distance to be covered Distance to
terminal Important
attributes

Gross tonnage
MLP neural
Previous port of call network

Accuracy
Next destination port Accuracy
calculation
calculation

Course & Heading


Vessel service
Current speed Current speed Compare
result

Change in speed over the last 3 hours


Average speed based on last 12
hours
Output Parameter Data
collection

preprocessing

 ATA [minutes]: The actual time of arrival of the vessel


Normalization MLP neural
network

 ATA=((D ATA - Dcollection )*1440)+((H ATA - Hcollection)*60)+(M ATA - Mcollection) Important


attributes

MLP neural

• Dcollection= Data Collection Day (mm/dd/yyyy) network

Accuracy
• D ATA = Arrival Day (mm/dd/yyyy) Accuracy
calculation
calculation

• Hcollection = Data Collection Hour (Hours)

1
• H ATA = Arrival Hour (Hours) Compare
result

• Mcollection = Data Collection Minute (Minutes)


• M ATA = Arrival Minute (Minutes)
Output Parameter Data
collection

preprocessing

 Delay [minutes]: calculated as the difference between


the actual time of arrival (ATA) and the ETA. Normalization MLP neural
network
• Expresses delay (+) and early(-) arrival
Important
attributes
DELAY = ATA - ETA
MLP neural
network
 Delay=((D ATA - DETA )*1440)+((H ATA - HETA )*60)+(M ATA - META )
Accuracy
Accuracy
calculation
calculation

• DETA = Estimated day of arrival (mm/dd/yyyy)

2
• D ATA = Arrival Day (mm/dd/yyyy) Compare
result
• HETA = Estimated hour of arrival (Hours)
• H ATA = Arrival Hour (Hours)
• META = Estimated minute of arrival (Minutes)
• M ATA = Arrival Minute (Minutes)
Data preprocessing Data
collection

preprocessing

 Delete missing data by eliminating records with


incomplete data Normalization MLP neural
network

 Remove. Out-layers (Using delay as parameter). Important


attributes

Delay<Q1-1.5X|Q3-Q1| or Delay>Q3 + 1.5|Q3-Q1| MLP neural


network

 Normalize data to (0,1)for accurate result using the Accuracy


calculation
Accuracy
calculation

equation
Compare
( X − X min) result
Xnew =
X max − X min
Important attributes Data
collection

 Identify the most important variables using the “Garson’s


Algorithm”
preprocessing

Hidden 1 Hidden 2 ….Hidden n Association MLP neural


Input 1 C1,1 C2,1 Cn,1 rule network
Input 2 C1,2 C2,2 Cn,2
Input 3 C1,3 C2,3 Cn,3 Important
…Input n C1,n C2,n Cn,n attributes

MLP neural
network
Relative contribution of
Contribution of each input neuron to output Accuracy
Accuracy
calculation
input neuron to output ri,j = Ci,j /∑ Ci,j calculation

(Ci,j )= Wi,j * Wo,i

Hidden 1 Hidden 2 ….Hidden n Hidden Hidden 2 ….Hidden n Sum Compare


result
Input 1 W1,1 W2,1 Wn,1 1
Input 2 W1,2 W2,2 Wn,2 Input 1 R1,1 R2,1 Rn,1 S1
Input 3 W1,3 W2,3 Wn,3 Input 2 R1,2 R2,2 Rn,2 S2
…Input n W1,n W2,n Wn,n Input 3 R1,3 R2,3 Rn,3 S3
Output Wo,1 Wo,2 Wo,n …Input n R1,n R2,n Rn,n Sn

Relative importance (Rj) = (Sj/∑ Sj)*100


Method Data
collection

preprocessing
 Neural Network
Association MLP neural
rule network
Overall ETA = ETA to Pilot Point + ETA to Terminal
Important
attributes

MLP neural
network

Accuracy
Accuracy
calculation
calculation

Compare
result
Results Data
collection

preprocessing

 To Pilot point To Terminal Association MLP neural


rule network

Important
attributes
Predicted Time of Arrival vs Actual Predicted Berth Arrival Vs Actual Berth
Arrival Time Arrival MLP neural
0.6 network
0.8 Actual time of arrival
Predicted Berth Arrival
0.5
Accuracy
predicted time of arrival Accuracy
calculation
0.6 Actual Berth Arrival calculation

Berth Arrival
0.4
Arrival

0.3
0.4
0.2 Compare
0.2 result
0.1

0 0
1 2 3 4 5 1 2 3 4 5 6

Vessel Vessel

MAE=0.086111 , RMSE=0.148524 MAE=0.176388488 , RMSE=0.208813


Results Data
collection

preprocessing

 To Pilot point To Terminal Association MLP neural


rule network

Important
attributes
Predicted Delay vs Actual Delay Predicted Berth Delay vs Actual Berth
0.5 0.9 Delay MLP neural
network
0.45 Predicted delay 0.8
Predicted Berth Delay
0.4
Actual delay 0.7 Accuracy
0.35 Actual Berth Delay Accuracy
calculation

Berth Delay
0.6 calculation
0.3
Delay

0.5
0.25
0.4
0.2
0.3
0.15
0.2 Compare
0.1 result
0.1
0.05
0 0
1 2 3 4 5 6 1 2 3 4 5 6

Vessel Vessel

MAE=0.062686499 , RMSE=0.062686 MAE=0.35621548 , RMSE=0.460897


Results Summary (Errors) Data
collection

preprocessing

Association MLP neural

To Pilot Point To Terminal Total Error


rule network

Important
attributes

MAE 0.086111 0.176388 0.262498


MLP neural
network

Accuracy
RMSE 0.148524 0.208813 0.357337 Accuracy
calculation
calculation

MAE 0.0626864 0.356215 0.4189014 Compare


result

RMSE 0.062686 0.460897 0.523583


Observation

 Using delay as an output variable performed better


than ATA from last port of call to the sea buoy

 Poor performance of delay between the buoy and


terminal was observed which is likely due to the
culture of the port :
• Day light restriction
• Dredging
• Pilot policies
• Restrictions due to vessel size
Acknowledgement

U.S. DOT Center


for Advanced Multimodal
Mobility Solutions and Education
Thank you!

Akintola Aremu
Graduate Student, Research Assistant
Dept. of Transportation Studies
Texas Southern University
Email: a.aremu3682@student.tsu.edu

You might also like