EFFICIENT ORGANIZATION OF TRUCK PLATOONS BY MEANS

OF DATA MINING

Application of the Data Mining Technique for the Planning and Organization

of Electronically Coupled Trucks

Ralph Kunze, Richard Ramakers, Klaus Henning and Sabina Jeschke

Center of Learning and Knowledge Management and Institute of Information Management in Mechanical Engineering

RWTH Aachen University, Dennewartstrasse 27, Aachen, Germany

Keywords:

Truck platoons, Platoon driving, Electronic coupling, Road capacity, Data mining, Sequential pattern mining.

Abstract:

One opportunity to manage the increase of freight transportation and to optimize utilization of motorway

capacities is the concept of truck platoons. With the aid of Advanced Driver Assistance Systems, trucks are

electronically coupled keeping very short gaps (approx. 10 meters) to form truck platoons on motorways.

This contributes to a couple of advantages, such as a reduction of the inter-vehicle distance, which leads to

an improved vehicle occupancy and gained road space, an optimization of trafﬁc ﬂow, the reduction of fuel

consumption advantaged by slipstream driving, a relief for professional drivers and ﬁnally an increase in safety

due to the longitudinal and lateral guidance. In this paper, a brief introduction into these truck platoons is given,

which have been realized within the project KONVOI. The paper focuses on the calculations for an efﬁcient

planning and organizing of truck platoons, which could have been realized by means of the application of data-

mining technique. Therefore, the efﬁciency criterion of electronically coupled truck platoons is discussed as

well as the framework of the data-mining technique and the algorithm for the calculation of efﬁcient truck

platoons are presented. Finally, the experimental results of the application of the data-mining technique and

their relevance for an efﬁcient operation of truck platoons is presented.

1 INTRODUCTION

1.1 Initial Situation

The integration of the new European member coun-

tries is a challenging component for national trafﬁc

planning in the near future. Especially countries with

a central geographic position within Europe, such as

Germany, have to shoulder the majority of the fu-

ture trafﬁc emergence. Additionally, this trafﬁc will

predominantly encumber the road. Due to the in-

crease of global freight transportation, the maximum

road capacity in several countries worldwide is nearly

reached (Economic and Social Commission for Asia

and the Paciﬁc, 2008). A modern national economy

needs an efﬁcient trafﬁc system to face such a chal-

lenge successfully. Otherwise, today’s even worse

trafﬁc situation will be pre-assigned to collapse.

The importance of this fact was emphasized by

studies from the European Commission in 2006

(Commission of the European Communities, 2006).

Between 1995 and 2004 a growth of 35% in road

freight transport was detected. Furthermore, between

the years 2000 and 2020 an increase of 55% in road

transportation is expected (Commission of the Euro-

pean Communities, 2006). In the year 2003, the Euro-

pean Commission stated that every day 7,500 kilome-

ters of the European road system are being blocked by

trafﬁc jams (Commision of the European Communi-

ties, 2003). Environmental pollution, safety risks and

a loss in efﬁciency for the economy are only some

of the effects that result from these factors. Similar

problems are known and discussed worldwide.

One possibility to face the rising trafﬁc volume

on the roads is the modal shift to other types of

transportation (e.g. rail, shipping). Further poten-

tial has the optimization of the road-side trafﬁc ﬂow

by driving assistance systems. Since the 90s, Ad-

vanced Driver Assistance Systems (ADAS) for trucks

have been on offer, including pre-adjustment of speed

and distance to the front vehicle. This is exerted

automatically via computerized engine- and brake-

management in connection with an automated trans-

mission. The combination of an Adaptive Cruise

104

Kunze R., Ramakers R., Henning K. and Jeschke S. (2010).

EFFICIENT ORGANIZATION OF TRUCK PLATOONS BY MEANS OF DATA MINING - Application of the Data Mining Technique for the Planning and

Organization of Electronically Coupled Trucks.

In Proceedings of the 7th International Conference on Informatics in Control, Automation and Robotics, pages 104-113

 SciTePress

Control (ACC) together with an Automatic Guidance

(AG) leads to autonomous driving. The difference be-

tween platooning and autonomous driving makes the

necessity of a leading vehicle. Following trucks can

go far distances without any manual engagement by

the driver as long as another ahead-driving vehicle ex-

ists. Nevertheless, each truck must be assigned with a

truck driver at all times due to legal rules and regula-

tions. Due to the platoons, smaller distances between

the vehicles (up to 10 meters) can be realized. These

truck platoons contribute to an optimization of trafﬁc

ﬂow up to 9% and a reduction of fuel consumption

due to slipstream driving (Savelsberg, 2005).

The development and evaluation of the practical

use of truck platoons is the objective of the project

KONVOI, which was funded by German’s Federal

Ministry of Economics and Technology. The Project

KONVOI is an interdisciplinary research project with

partners of RWTH Aachen University, industry and

public institutions, which ended after a duration of 49

months with test runs on German highways at the end

of May 2009 (Figure 1). With the assistance of virtual

and practical driving tests by using experimental ve-

hicles and a truck driving simulator, the consequences

and effects on the human-, the organization- and the

technology-dimension have been analyzed (Henning

et al., 2007).

Figure 1: Test Run on German Motorways (March 2009).

1.2 Problem Deﬁnition

For planning and organization of such truck platoons,

as well as evaluation and simulation purposes, it is

necessary to implement an algorithm which searches

for economic truck platoons within the planned or

current driven routes of all participating trucks. In

any case of organized platoon driving, it is essential

to search for and group possible participants. In prin-

ciple, the ”common route” is one general search crite-

rion, whereas other criteria (e.g. proﬁt, waiting-time,

properties of the participating vehicles) can differ be-

cause of the focus.

In this paper, a data-mining technique is presented

to solve the mentioned problem of searching for pos-

sible platoons which satisfy given speciﬁc criteria (in

the case of truck platoons: economy). The scenario of

driver organized truck platoons, on which KONVOI

is based, is presented in section 2. Section 3 explains

the operational and organizational structure of platoon

systems. Before introducing the so called TPSpan-

Algorithm and related work as well as the possibil-

ity to use this algorithm as a solution for the problem

of planning and organizing truck platoons in section

5, the criterion of a efﬁcient usage of electronically

coupled trucks as well as the calculation basis for the

efﬁciency of truck platoons are deﬁned in section 4.

Finally, in section 6 the experimental results of the

application of the data-mining technique is presented.

2 THE SCENARIO ”DRIVER

ORGANIZED TRUCK

PLATOONS”

The project KONVOI is based on the scenario ”Driver

Organized Truck Platoons” (Figure 2) which was de-

veloped in the project ”Operation-Scenarios for Ad-

vanced Driving Assistance Systems in Freight Trans-

portation and Their Validation” (EFAS) (Henning and

Preuschoff, 2003). In the scenario ”Driver Orga-

nized Truck Platoons”, the platoons can operate on

today’s existing motorways without extending the in-

frastructure and the driver has the permanent con-

trol of the autonomous driving procedures (Henning

and Preuschoff, 2003). The creation of a platoon de-

pends on the initiating driver who delivers the neces-

sary data about time and place of meeting, the des-

tination, as well as the required truck telemetric data

(loading weight, engine power etc.) with the help of a

Driver Information System (DIS). The high ﬂexibility

of truck transportation is not lost, because schedul-

ing, like in rail trafﬁc, is dispensable. After activat-

ing the ADAS, a selection of the best matching pla-

toons is automatically shown. The ADAS informs the

driver and prepares the participation to the selected

platoon. The DIS acts as a human machine interface

of the platoon system and helps the truck driver to

plan the route and guides the driver to the meeting

point (Friedrichs et al., 2008).

The driver has to initialize and respectively con-

ﬁrm all of the platoon maneuvers in order to build

and to dissolve the platoon. As soon as the ﬁnal po-

sition in the platoon is reached, an automated longi-

EFFICIENT ORGANIZATION OF TRUCK PLATOONS BY MEANS OF DATA MINING - Application of the Data

Mining Technique for the Planning and Organization of Electronically Coupled Trucks

105

Table 1: Reduction of Fuel Consumption based on Theory, Simulation and Test (Bonnet and Fritz, 2002).

THEORETICAL SIMULATION MEASUREMENT

DAIMLER DAIMLER

Fuel Consumption (1st vehicle) 2,17% (14t); 1,64% (28t) 2% (28t) 6% (14t)

Fuel Consumption (2nd vehicle) 38,06% (14t); 28,76% (28t) 19% (28t) 21% (28t)

Table 2: Reduction of Fuel Consumption due to Electroni-

cally Coupled Platoon Driving.

PLATOON FUEL

STRUCTURE REDUCTION

1st Platoon Vehicle 2%

(Leading Vehicle)

2nd Platoon Vehicle 11%

(Following Vehicle)

3rd Platoon Vehicle 13%

(Following Vehicle)

Average Fuel Reduction, 6.50%

2 Vehicles

Average Fuel Reduction, 8.67%

3 Vehicles

Average Fuel Reduction, 9.75%

4 Vehicles

Average Fuel Reduction, 10.40%

5 Vehicles

meeting point or reduce speed until they got together.

Another criterion for extra personnel costs is the pla-

toon driving itself and the associated platoon maneu-

vers (e.g. a platoon has to be dissolved at a working

area). It is also possible that a truck, which partic-

ipates in a platoon, has to drive with reduced speed

because of another, slower participant (e.g. uphill).

Due to the fact that the extra personnel costs

caused by platoon maneuvers or speed reduction be-

cause of slow participants are unascertainable without

accurate road and trafﬁc information, therefore those

inﬂuencing criteria will be ignored. The only criteria

which take account of extra personnel costs will be

the waiting time Time

(l) for each participating truck

i and the wage rate k

Costs

(l) = k

· Time

(l) (6)

4.4.3 Boundaries of the Efﬁciency of Truck

Platoons

To get a statement on the maximum of waiting time,

equation 3 has to be transformed and complemented

by equations 4 and 6. The efﬁcient criterion of

electronically coupled truck platoons is originated in

equation 7.

Length(l)

100

· ∆B

e,Size(l)

> Time

(l) (7)

With this efﬁcient criterion it can be easily

proofed, if the beneﬁt of the operation of electroni-

cally coupled truck platoons is bigger than their costs.

5 DATA MINING TECHNIQUE

5.1 Related Work

Discovering patterns in sequences is an important

knowledge discovery and data-mining research area.

There are many different interesting ﬁelds like discov-

ering rules in so called events (Mannila et al., 1997)

or ﬁnding text subsequences (Califano and Rigout-

sos, 1993) that match a regular expression. In 1995,

Agrawal and Srikant introduced the problem of min-

ing sequential patterns (Agrawal and Srikant, 1995):

”A database with customer transactions is given.

Each transaction consists of the ﬁelds; customer-id,

transaction-time and a set of items purchased in this

transaction. Quantities of items purchased in a trans-

action are not considered. The problem of mining

sequential patterns is to ﬁnd all sequences that have

a certain user-speciﬁed minimal support.” This dif-

ﬁculty deals with the questions; ”which items are

bought in which order” or ”which item will be bought

next”. Agrawal and Srikant presented three algo-

rithms for solving the sequential pattern problem and

many other algorithms have been proposed to speed

up the mining process (e.g. (Zaki, 2001); (Pei et al.,

2004)). In general, those algorithms are based on

two different methods for mining sequential patterns:

apriori-based methods and pattern-growth methods

(Antunes and Oliveira, 2004).

The apriori-based methods are, as the name states,

based on the so called apriori condition. This con-

dition is used to generate larger candidates based on

smaller frequent sequences in the so called join-step.

Each generated candidate is then validated in the fol-

lowing so called prune-step. These two steps are

repeated until there are no more frequent sequences

found. The pattern-growth methods deduce from a

smaller frequent sequence, a longer sequence. Instead

of generating candidates, these methods directly look

for frequent sequence.

In 2003, Yan et al. introduced the problem of min-

ing closed sequential patterns which means that in-

EFFICIENT ORGANIZATION OF TRUCK PLATOONS BY MEANS OF DATA MINING - Application of the Data

Mining Technique for the Planning and Organization of Electronically Coupled Trucks

109

stead of ﬁnding all frequent sequences, only the ”most

supported” sequences are searched for (Yan et al.,

2003). They presented a new algorithm to ﬁnd closed

sequential pattern, called CloSpan, which is based on

a pattern-growth method introduced in 2001 by Pei

et al. Figure 6 illustrates the main idea of the ap-

proach. The items are nodes in a lexicographical se-

quence tree. Each path of the tree is a frequent se-

quence found in the database. So called projected

databases are stored for each node of the tree. If the

projected databases of two different paths are equal

to each other, the mining for more frequent sequences

can be eventually stopped. In ﬁgure 6, the mining for

further frequent sequences can be stopped at node B

in the path B because the projected database of this

node is equal to the projected database of node B in

the path A, B.

Figure 6: Illustration of the Algorithm to Find Closed Se-

quential Pattern.

The problem of mining frequent sub-routes or

economic truck platoons cannot be solved easily by

any of those algorithms. Due to the small support of

two and the given sequences of hundreds of sections,

those algorithms collapse. In the case of platoons the

support has to be two, because two is the smallest

number of participants to form a platoon. However,

these algorithms are good approaches in ﬁnding pos-

sible solutions for solving this problem.

5.2 The Truck Platoon Sequential

Pattern Algorithm

The so called TPSpan-Algorithm (Mining Truck Pla-

toon Sequential pattern) had been already presented

in detail in 2008 (Meisen et al., 2008). In this section,

we will give a brief introduction into the TPSpan-

Algorithm which searches for possible platoons in a

route database, especially optimized to ﬁnd economic

truck platoons. The algorithm is split in four phases

and illustrated in Figure 7.

In the ﬁrst phase, the exclusion criteria are used to

eliminate trucks or routes which do not meet those cri-

teria. Within the second phase, the grouping criteria

are applied to secure the forming of platoons within

similar conditions. In the third phase, the data-mining

technique checks each group for possible economic

truck platoons regarding the assessment criteria. Fi-

nally, in the fourth and last phase, the economic truck

platoons are grouped by truck-id. The grouped infor-

mation is then transmitted to each truck (e.g. meeting

points, proﬁt).

Figure 7: Illustration of the Algorithm to Find Truck Pla-

toon Sequential Pattern.

The used data-mining technique in the third phase

is based on the projected pattern-growth idea pre-

sented by Pei et al. (2001) and has been enhanced for

solving the problem of mining frequent sub-routes.

It generates a lexicographic tree as shown in Figure

6. Instead of starting with each frequent section, the

tree starts with those frequent sections which have a

meeting opportunity (Meet(s

) = 1 ). This means that

the ﬁrst section has to be e.g. a rest area. Due to

this, the width of the tree is reduced and the condi-

tion for a platoon (starting with a meeting opportu-

nity) is ensured. Another important improvement is

the so called node-compress-method which is used to

reduce the validation of the assessment criteria. In-

stead of validating the criteria for every platoon, the

common distance of a platoon will be increased as

long as the number of participants does not change. It

is ensured that the algorithm will still search for the

most economic platoons: A platoon l

= (s

, . . . , s

)

with Size(l

) = k is always more economical than a

platoon l

= (s

, . . . , s

) with Size(l

) = k, whereas

Length(l

) > Length(l

Figure 8 shows the framework of the data mining

technique. The algorithm is working from the root

of the tree to the leaves. In each node, the algorithm

calculates possible platoons (line 4) and determines

new children (line 6). If the projected database of

the node does not support any more platoons (line 1)

the recursion terminates. The framework also shows

the already mentioned node-compress-method (line

3) which returns a true value if the compression termi-

nated with a split. If this is the case, further children

are possible, other wise the routes of the projected

database end and no more children can be determined

(line 5).

ICINCO 2010 - 7th International Conference on Informatics in Control, Automation and Robotics

110

7 CONCLUSIONS

In this paper, we have introduced a data mining tech-

nique to plan and organize platoons. Furthermore,

we have introduced and presented experimental re-

sults for an application area ”truck platoons: trains

on road”. For this case, we have examined that it is

possible to ﬁnd truck platoons and that the amount

of platoons increase exponentially with the amount of

routes (participating trucks). Due to this rise, pruning

parameters are necessary. Further experiments have

shown that it is possible to ﬁnd truck platoons reli-

able, efﬁcient and ﬂexible, even if pruning parameters

are used. We have given suggestions for these param-

eters to achieve the mentioned factors. Further work

has to be done in the ﬁeld of realization. We have

presented the dilemma of different results of reduc-

ing fuel consumption. Additional effort has to be put

into the goal closing the gap between results of fuel

reduction based on theory, simulation and tests and

thus provide a further substantial contribution to the

efﬁciency of truck platoons.

REFERENCES

Agrawal, R. and Srikant, R. (1995). Mining Sequential Pat-

terns. In Proc. of the 11th Intl Conf. on Data Engi-

neering, Taipei (Taiwan).

Antunes, C. and Oliveira, A. L. (2004). Sequential Pattern

Mining Algorithms: Trade-offs be-tween Speed and

Memory. In Proc. of the 2nd Intl Workshop on Mining

Graphs, Trees and Sequences (ECML/PKDD).

Bonnet, C. and Fritz, H. (2002). Fuel Consumption Reduc-

tion Experienced By Two PRO-MOTE-CHAUFFEUR

Trucks. In Electronic Towbar Operation.

Califano, A. and Rigoutsos, I. (1993). Flash: A fast look-

up algorithm for string homology. In Proc. of the 1st

Intl Conf. on Intelligent Systems for Molecular Biol-

ogy, Bethesda (USA).

Commision of the European Communities, E. U. (2003).

Europe at a crossroad - The need for sustainable

transport. Brussels, Belgium.

Commission of the European Communities, E. U. (2006).

Keep Europe moving Sustainable mobility for our

continent. Brussels, Belgium.

Economic and Social Commission for Asia and the Paciﬁc,

U. N. (2008). Statistical Yearbook for Asia and the Pa-

ciﬁc. United Nations Publication, Bankok, Thailand.

Friedrichs, A. (2008). A Driver Information System for

Truck Platoons. VDI, Duesseldorf, Germany.

Friedrichs, A., Meisen, P., and Henning, K. (19.-

22.05.2008). A Generic Software Architecture for a

Driver Information System to Organize and Operate

Truck Platoons. International Conference on Heavy

Vehicles (HHVT2008), Paris, France.

Henning, K. and Preuschoff, E. (2003). Einsatzszenar-

ien fuer Fahrerassistenzsysteme im Strassengueter-

verkehr und deren Bewertung. VDI, Duesseldorf, Ger-

many.

Henning, K., Wallentowitz, H., and Abel, D. (2007).

Das Lkw-Konvoisystem aus den Perspektiven

Informations-, Fahrzeug- und Automatisierungstech-

nik. In Mechatronik 2007 - Innovative Produkten-

twicklung.

Kunze, R., Tummel, C., and Henning, K. (19.-20.12.2009).

Determination of the Order of Electronically Coupled

Trucks on German Motorways. 2nd Conference on

Power Electronics and Intelligent Transportation Sys-

tem (PEITS 2009), Shenzhen, China.

Mannila, H., Toivonen, H., and Verkamo, A. I. (1997). Dis-

covery of frequent Episodes in Event Sequences. In

Data Mining and Knowledge Discovery.

Meisen, P. (2006). Data-Mining Technique for Truck

Platoons. RWTH Aachen University, ZLW/IMA,

Aachen, Germany.

Meisen, P., Henning, K., and Seidl, T. (2008). A Data-

Mining Technique for the Planning and Organization

of Truck Platoons. In Proceedings of the International

Conference on Heavy Vehicles.

Pei, J., Han, J., Mortazavi-Asl, B., Pinto, H., Chen, Q.,

Dayal, U., and Hsu, M.-C. (2004). PreﬁxSpan: Min-

ing Sequential Patterns Efﬁciently by PreﬁxProjected

Pattern Growth. In Proc. of the Intl. Conf. Data Engi-

neering (ICDE’01).

Savelsberg, E. (2005). Lastenheft fuer elektronisch gekop-

pelte Lkw-Konvois. VDI, Duesseldorf, Germany.

Schmitz, J. (2004). Energy-economic cost-beneﬁt-analysis

of electronically coupled truck convoys. RWTH

Aachen University, ZLW/IMA, Aachen, Germany.

Yan, X., Han, J., and Afshar, R. (2003). CloSpan: Min-

ing Closed Sequential Patterns in Large Datasets. In

SDM’03.

Zaki, M. J. (2001). SPADE: An Efﬁcient Algorithm

for Mining Frequent Sequences. In Proc. of Ma-

chine Learning Journal, Special issue on Unsuper-

vised Learning (Doug Fisher, ed.).

EFFICIENT ORGANIZATION OF TRUCK PLATOONS BY MEANS OF DATA MINING - Application of the Data

Mining Technique for the Planning and Organization of Electronically Coupled Trucks

113