ELECTRIC VEHICLE ROUTE RECOMMENDER SYSTEM

João C. Ferreira

, Vitor Monteiro

and João L. Afonso

Centro Algoritmi, Univ. of Minho, Guimarães, Portugal

ADEETC, ISEL, Lisboa, Portugal

Keywords: Electric Vehicle, Recomender System, Range Prediction, Driver Profile.

Abstract: This paper presents a recommender system responsible for processing information that will help the driver

in the daily use of his Electric Vehicle (EV), minimizing the problem of range anxiety through a

personalized range prediction and by presenting in real time relevant information about the charging stations

that can be reached within the range autonomy. Given the success of recommendation systems on automatic

delivery of relevant information in numerous areas of usage, this type of systems can also be applied in the

electric mobility scenario, with the objective of maximizing the relevance of the information presented to

the driver, which should be the strictly needed data for the driver to make important decisions, filtering out

the unnecessary information.

1 INTRODUCTION

Aiming the electric mobility in Smart Grids (SG),

mainly electric vehicles and the driver habits, in this

paper is proposed a recommendation system that

aims to present to the users the recommended

charging point according to their preferences, with

their usual choices and according to the data

collected from public transportation, traffic

information and charging places. The

recommendation system is also used to present to

the user sites and services near of the charging point

or typical services associated with mobility in large

urban areas. The system communicates with the

vehicle system to obtain information of the battery

status and the range prediction estimation (Ferreira,

2011). The integration of information

recommendation systems with information of the

Smart Grids energy systems in conjunction with the

interface with the EV allows deliver alerts to the

driver during the normal vehicle use. Control the

vehicle's range and indicating the nearest point of

charging through geographical positioning

information, use the user profile to indicate the

direction of the planned route, loading preferences

between fast or slow charge, identifying the

proximity to public transportation scheme, are the

main objectives of the proposed recommendation

system. The system provides to the driver several

points of interest (e.g. loading points, interface with

public transport, parking lots with charging or tourist

areas of interest), based on a recommendation

system, based on the profile and the preferences of

the driver. The system will target a mobile device

that may be a Personal Data Assistant (PDA) or a

device integrated into the EV, taking into account

the specifics of these devices which have advantages

such as mobility and integration with geographic

positioning systems and disadvantages such as

limitation of processing capacity and size of the

graphic interface.

2 RECOMENDER ENGINE AND

SYSTEM

Given the success of recommendation systems in

automatically delivering the relevant information in

numerous areas of usage, it can be applied in this

scenario as well, with the objective of maximizing

the relevance of the information presented to the

driver, which should be the strictly needed for

him/her to make decisions, filtering out the

unnecessary one.

Based on the available information (charging

points, traffic and public transportation information,

and EV range autonomy), the recommendation

system will choose (recommend) the ‘right’

information, based on user pre-defined preferences

expressed in their profile. This system uses n

C. Ferreira J., Monteiro V. and Afonso J..

ELECTRIC VEHICLE ROUTE RECOMMENDER SYSTEM.

DOI: 10.5220/0003954400490058

In Proceedings of the 1st International Conference on Smart Grids and Green IT Systems (SMARTGREENS-2012), pages 49-58

ISBN: 978-989-8565-09-9

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

dimensions to calculate utility function (information

relevance measurement) for all input information:

user D

, information D

and context D

. A diversity

of recommendations is displayed. So, a

recommended function has a new dimension d

(means context), and is a three dimensional vector u

(user, item, and context), see Figure 1:

Figure 1: Overview of the Recommender System.

Input Information: Is the available information

and represents charging places and other information

related to EV driver.

Best Path: Is the algorithm implemented to

define the best travel to be performed.

Driver Profile: Represents past user options in a

three dimensional matrix based on selected

information. In our implementation this field has

only two values: value 1 represents a user choice,

and value 0 signifies user disagreement. It is used to

improve recommender system performance, based

on the fact that user preferences change based on the

context. For example, in a tourism recommender

system, the year period changes the user’s

preferences, among others. Several authors,

(Gediminas, 2005) and (Tuzhilin, 2001), suggest that

context information is used based on a bi-

dimensional space, (User, Information), in a

multidimensional matrix, u (utility): u:D

x …

ÆR. Each D dimension is represented by an

information vector and the algorithm to find the best

tuple (d1,…,dn), that maximizes the utility function

(Gediminas, 2005). Our implementation uses as

context information based on four main types:

(1) function mode: “leisure” or “direct”;

(2) day period: “morning”, “afternoon” or

“night”;

(3) destination: “home” or “work”;

(4) trip status: “starting” “middle” or “ending”.

This information is stored on information

repository and is obtained by interaction with the

driver, (1) plus (3), and (2) plus (4), by the system

based on pre-defined rules. So, day period is pre-

defined on system, morning is until 12 am, afternoon

from 12 am to 19 pm, and night in the remaining

hours. Trip status is based on target distance, the

first 25% km is considered starting and the last

25% km are considered ending.

Recommender Information: Generates new

system output information, ordered by a utility

function. In this pre-release phase the proposed

system takes care of the charging point.

Each information type is represented by a vector,

with class values. The PCSs (Public Charging

Stations) are based on:

• Charging Modes: Slow, Fast;

• Operator: “MOBI.E”, “Better-Place”, “EDP”;

• Location Type: “Street”, “Small Station” and

“Big Station”;

• Services: “Without Services”, “Reservation”,

“Washing Services” and “Others”:

• Public Transportation Interface: “1”,”0”;

• Nearby Points of Interest (POI): “1”,”0” – only

if operator mode is on travel mode.

Utility (u) Calculation

User recommendation is based on utility function

(u), where only three items are showed to the user

(this is a configurable parameter). After distance

measure is introduced (Manhattan distance, others

could be implemented), it is calculated the utility

based on context information, for each piece of input

information, item I

: I

= {I

, I

, C

Where {I

, I

} are the characteristics of input

information I

and {C

} are the context

characteristics.

The utility function is based on the nearest

neighbor algorithm, see Figure 2, where it includes

context and distance function. For details see

(Pereira, 2010).

Figure 2: Nearest Neighbour Algorithm used to calculate

the function utility.

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

3 DRIVER PROFILE

Initial driver profile is manually created by the

driver, with the following information (Table 1):

Table 1: Driver profile of an Electric Vehicle (EV).

Propriety Description

User login

information

User name and password

Home Address GPS position of home address

Work Address GPS position of Drivers work

Car Information

Model, Year, Battery type and

power

Trip information

Work days or week-end +

holidays, start time, finish time,

distance (km), SOC level

This profile, later receives information about

driver trip (time, duration and km travelled) from the

tracking system. A resume of EV parameters (speed,

SOC level, travelled distance) is also stored and

associated with this profile for later range prediction.

Driving range (see section 4) is intensively

related to the driving style or mode. This happens in

all types of vehicles, but on EV, due to the weakness

related to the amount of energy stored on-board, this

relation is much clearer. Thus, changing driving

style and driving habits may be a considerable factor

on energy saving and on extending vehicle

autonomy. Considering the actions or driving habits

that can bring significant energy saving to the

vehicle operation, it is important to evaluate how

receptive EV drivers will be in changing their

driving style and driving habits towards the

achievement of the intended energy saving. The

study of this receptivity for change will be addressed

through a set of surveys made over a group of

potential EV drivers with different profiles, from

gender, age, education, job, etc. These surveys will

allow the identification of correlations between the

driver profile and new driving habits. Under these

surveys another factor that will be evaluated relates

with the acceptance for automated actions performed

by the vehicle control system. This type of actions

may be accepted for some of the driving process but

not for all driving process. For example, one driver

may accept the vehicle control system to

automatically turn off the air-conditioning under

certain conditions, but will not accept the system to

limit his driving speed. Part of this study will create

and store for further analysis a Driving Profile.

Driver profiles will play an important role to

determine the EV range, since range prediction will

be based on the assessment of the drivers’ usual

behaviour. An initial Driver profile can be identified

by the system, after being created manually by the

driver with the information presented in Table 1.

Part of this driver profile will be based on a

tracking application running on a driver’s mobile

device. This application will update driver profile

with travel distances, time and SOC levels, weather

information, etc. as showed in Figure 3.

Figure 3: Main module of the driver’s tracking system in a

mobile device with GPS and with information created

from the Drivers Movements database.

4 RANGE PREDICTION

Range prediction is the process of estimation the EV

range based on three main dependency types:

1) The EV with its main variables: the model of

the vehicle (mainly its performance under different

scenarios, speed, and acceleration), the chemical

technology of the batteries (as lithium-iron-

phosphate, lithium-titanate, or nickel-metal-hidride)

the batteries characteristics (mainly, variation of

SOC, temperature, lifespan, performance, specific

power, specific energy, and safety), and the EV

powertrain (electric motor and its power converter,

as well as the other electric parts, as batteries

charger, controllers, and power cables). This data is

stored on an information repository on a central

server. All of these parts will influence the SOC and

consequently the range prediction. The batteries

SOC, and other relevant parameters, are provided to

the main control system through CAN-bus

communication, and then these information’s are

stored in a data base (DB), in order to predict the

available range.

2) The driver behavior: speed and acceleration

information are taken from EV through the CAN-

bus communication, and the driver past behavior

ELECTRICVEHICLEROUTERECOMMENDERSYSTEM

(e.g., SOC level versus travelled distance achieved),

are stored in a DB. Weight is a manual input, and

driving directions are acquired based on the GPS

information.

3) Environment: current location, traffic

conditions (taken from a web service), road

information (in a distance graph), weather

information (wind and temperature, taken from a

web service), and altitude, taken from GPS.

The range prediction procedure is in a first step

based on the batteries SOC level, and using driving

behavior (relation of SOC level versus travelled

distance achieved, from past experience, stored in a

central information repository). This distance is

tuned based on weather information: if it is hot, a

percentage of the energy stored in the batteries is

taken for air-conditioning; if it is raining, a

percentage of energy is taken for the window

cleaning process. For night drive, also a percentage

of energy is taken for light services. A web service

brings traffic information, and based on past

experience (e.g., information about driving times

and traffic information), a driving range is predicted.

Current driving behavior (e.g., driving speed and

accelerations) are taking into account in this process.

Once it is achieved the EV range estimation, it is

started the calculation of the possible reachable

points based on current position.

Most of these estimation processes were

performed based on simple heuristics

(Ferreira, 2011). For example, if the weight is above

a configurable threshold, a percentage of reduction

is applied on the range (this value should be tuned

from past experience data and depending on EV

parameters). Also altitude was taken into

consideration, using Google maps to obtain the

altimetry of the desirable path, again several levels

were configured, and also is applied a percentage of

reduction in the range, in function of the altitudes to

overcome in the path. Temperature is an important

parameter, because of the relation with air-

conditioning. The temperature was divided in five

classes range: (1) less than 5ºC; (2) from 5ºC to

15ºC; (3) from 15ºC to 25ºC; (4) from 25ºC to 30ºC;

and (5) above 30ºC. For classes 1 and 5 it was

assumed that all drivers use air-conditioning, for

class 3 no usage is performed, and for classes 2 and

4 it is considered that a percentage of drivers use air-

conditioning (initial profile parameter, but past data

can tune this behavior). Traffic information is used

again as a parameter that can reduce range, because

possible starts/stops on traffic jams increase

consumption.

4.1 Range Representation

Once a range prediction is achieved, a topographical

search starts with the current driver position, based

in Figure 4. Main road nodes are used to check

distances from current position and a polygon

representation is achieved (see Figure 5 and Figure

6) based on Google API usage. A zone of

uncertainty can be marked, based on the uncertainty

parameters used to estimate the drive range (see

Figure 7).

Figure 4: Range representation process using Google

Maps API.

If the SOC level is below 25 % (available range

should be around 30 km-40 km) it is calculated

every road option with guidance to the nearest

charging point. Taking into account Figure 5, it was

considered Lisbon as the starting point. Since the

available range for the EV is around 160 km, the

implemented process starts to look for main

destinations in a radius of 130 km to 160 km. This

distance calculation is based on GPS coordinates of

correspondent places. For Lisbon as the starting

point, the process identified the following cities

(Figure 4, process (2)): Pombal, Leiria, Marinha

Grande, Ourem, Tomar, Évora, Grandola, Santiago

do Cacém and Sines. Then, the distances are

calculated based on Google Maps query (Figure 4,

process (3)), and the process identifies that Pombal

are out of the EV range. The distances calculated to

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

the other locations are within the available range of

the EV. For example, the distance from Lisbon to

Évora is 134 km, so the process (4) (Figure 4) looks

nearby villas, and process (5) (Figure 4), identifies

the ‘real’ distance. In the case of the present

example (Figure 6) and using the city of Évora as

destination, it is available more 26 km, which allows

increasing the range representation around Évora

with a radius of 20 km. The output of this iterative

process is represented in Figure 6. For every 5 km of

EV movement this map is again calculated and

represented. The web range estimator represents

range by the connection of main distances and

putting the polygon together. To do so, our

application uses Google maps API and shows the

polygon on a mobile device display, as showed in

Figure 5 and Figure 6.

Figure 5: Range estimation of a trip from Lisbon to north,

where four different cases are shown.

For the charging process, the range prediction

and their representation is performed in the same

way. Based on the charging level (SOC information)

the application predicts the range based on previous

driving parameters (past relations of SOC levels and

distances achieved stored in driver profile) and

based on this information represents using Google

Maps the regions that is possible to reach with that

charging level. System is prepared to generate alerts

about charging levels needed to reach a charging

station (it is assumed that a charging process is

always performed in a charging station, in the

driver’s home or in the work place). The range

prediction process has several uncertainty factors

that reflect driving behaviour and external condition

(e.g . traffic, road topology and weather). These

Figure 6: Representation of the charging range for

different SOC levels, with a charging process performed

in a public charging station in the centre of Porto city. On

left, with a low SOC level, a small area is achieved. On

right, with almost 90% of SOC level, a large area is

achieved.

Figure 7: Range estimation based on the uncertainty

factors shown in Figure 4.

factors showed in Figure 4 can be used to estimate a

safe range (green shadow in Figure 7) and a

maximum range. The red shadow in Figure 7 is

arange that is possible to achieve but the driver

needs to perform driving optimization (air condition

off and avoid big accelerations). This could be

helpful information because driver can customize his

behavior function of the range it needs to achieve in

their trip. This process is can be continue updated

and when SOC level is low this uncertain gets low.

ELECTRICVEHICLEROUTERECOMMENDERSYSTEM

5 REAL TIME INFORMATION

AND BEST PATH

Real time information is available from road

concessionaries, but outside these organizations the

access to this information is most of times denied.

There are technical issues to solve because each

organization has its own data format, and data

integration is a real problem. During current

research work several approaches were performed to

have access to these data (with the Municipal

Chambers of Lisbon and Loures, and with the

company Brisa), but all the requests have been

denied. The problem was contoured with the

creation of a web crawler to pick traffic information

from specialized sites by pre-defined heuristics, and

a XML file with traffic information was created

(Fontes, 2008). The XML file is an approach of

future data integration from different source

providers. In Figure 8 is illustrated the process of

XML file with road information oriented to a geo-

reference graph.

Figure 8: Semi-automatic creation of a XML file with road

information to be used for graph creation.

The Crawler (WebNews, version 1.0) was

configured to pick traffic information from TVI web

site [http://www.tvi.iol.pt/transito.php] and also a

Web service from Sapo was implemented.

Information about nodes were checked against a

heuristic table, where a conversion factor (CVF)

reduces the node speed traffic (if the rode is blocked,

with no traffic flow, the CVF is zero). Figure 9

illustrates this process. The result is stored in an

adjacency matrix, where the number aij represents

the cost of going from i to j. This process creates a

graph representing the map, where the arcs represent

roads and nodes represent intersections or traffic

areas. To determine the best path between two points

on a map an algorithm could be applied to this graph

to find the path with less weight between the two

desired points.

The weight of an arc is basically the average

time in seconds that it needs to be traveled, for this it

is used the equation:

Time = (L / V) * 3600,

where L represents the size of the graph in km

and V is the top speed allowed. For all arcs there is a

speed limit which serves as the basis for the weight

of the cases in which there is no traffic information.

When there is traffic information for a particular arc,

the weight of this arc is affected because the traffic

influences the speed. For example, if an arc with

heavy traffic has a speed limit of 90 km/h, given the

existing traffic, the reduction factor of the maximum

speed has a value set between 0 and 1, based on

current traffic information, causing the increase of

the time required for this arc to be traversed, as

much as its weight increases. If the transit is cut off,

the factor of speed reduction assumes the value 0,

and the maximum attainable speed is set to 0 km/h,

what causes this arc to have an infinite weight.

If traffic is proceeding smoothly, without any

problems, the factor of speed reduction has a value

equal to 1, and the maximum attainable speed

becomes equal to the speed limit, which makes this

arc to assume the lowest weight possible, what is

translated in a reduction in the mean time the arc

needs to be traversed. This weight can integrate also

public transportation information and a price

associated with CO

emission of private

transportation. Users can choose the impact

parameters based on their strategies, in order to save

money, to satisfy conveniences, or even to save

time. For a complete description see (Fontes, 2008).

Figure 9: Graph node weight calculation based on traffic

information from web sites (traffic information in

Portuguese language).

Example: Path A1 (Alhandra – Alverca).

Maximum speed allowed on this section: 120 km/h

(motorway), but traffic information shows average

speed is 84 km/h. Distance is 5km, so

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

arc weight = (arc distance / average speed) x 3600 s

= (5000/84000) x 3600 ≈ 214 s.

Let’s consider, in this example, that public

transportation takes more or less twice the time

spent with private transportation (by car), consisting

in a time of 400 s.

Transportation price is 1 € and car transportation

is 2 €, including the price for CO

emissions and fuel

price. If drivers choose the same weight factor for

time and price, then the arc weight for car

transportation is 214 x 2 = 428, and the

transportation arc weight is 400 x 1 = 400. In this

case these values approximately the same, but if the

process takes care of parking price of the vehicle in

the city, the system increases again the arc weight

for private car, and then the system suggests the

option of public transportation (PT), showing the

next PT arriving to the nearest PT stop. System is

able to provide orientation to the nearest PT stop.

To determine the best route or the quickest route

was implemented a Dijkstra's algorithm, where its

running time is proportional to N

, with N being the

number of nodes in the graph. The graph to be used

by this Dijkstra's algorithm is represented by an xml

file. The best way to apprehend the quickest way is

determined using the Dijkstra algorithm. In our web

application there is a class called Dijkstra, which is

where the algorithm is implemented. For a complete

description method sees (Fontes, 2008). Information

exchanges with external systems are based on XML

files.

6 CASE STUDY

EV create new paradigms to mobility due to its

limited range and long recharging times, and

inevitably have impact on the day-to-day of the

drivers, who have to plan their journeys and stops

for recharging. These problems will tend to decrease

with the technological advances that increase the

autonomy of the vehicles, and with the development

of a network of battery charging places, with an

increasing number of alternative sites, as well as

with a larger geographical dispersion.

Current work shows scenarios that could be

considered as representative for most types of

electric vehicles in the near future. In Table 2 are

presented the data used in the case study.

Table 2: Data used in the case study.

Characteristics Values

EV Autonomy 150 km

Charging Points in Portugal 1,300

Slow Charging 6 Hours

Fast Charging 30 Minutes

Change Battery 10 Minutes

6.1 Case 1 - Day-to-day Urban

This scenario shows the use of Electric Vehicle (EV)

by a driver who lives on the outskirts of a big city,

and uses the EV to go to the work. The driver takes

the kids to school every day, and after the job, goes

to a gym session before making the return trip. His

typical driving day is:

• 07:30 - He leaves home with the EV fully

charged, after charging during the night;

• 08:15 - He covers 30 km until reaches the school

of the first son;

• 08:30 - He traverses 5 km to reach the school of

the second child;

• 09:00 - He covers 10 km to came to his work

place;

• 12:30 - He goes out for lunch in his EV covering

a total of 10 km;

• 18:00 - He covers 10 km to the Gymnasium;

• 19:30 - He covers 10 km to the school of the

child;

• 19:45 - He covers 5 km to the school of the other

son;

• 20:30 - He covers 30 km in the way back home.

The driver returns home with 25% of the EV

battery capacity.

This is the optimal scenario, appointed by

opinion builders to justify the viability of EV,

vehicles even with the current state of autonomy.

However in this scenario an extra utilization of the

EV cannot exist, as leaving home in the evening for

dinner, or visiting friends. Besides, in this scenario

all EV owners have to be able to charge their

vehicles at home overnight. For the vast majority of

the population that lives in metropolitan areas, it is

not possible to charge the batteries of the vehicles

during the night, since the vast majority of them live

in buildings without parking or garage that make

possible the overnight charging. For these cases the

batteries charging must be carried out during the

day, or in the middle of a travel, as with today's

vehicles powered by fossil fuels.

ELECTRICVEHICLEROUTERECOMMENDERSYSTEM

Thus, maintaining the same schedule, but

removing the batteries charging during the night, the

driver will enter the vehicle with 25% of the battery

capacity. Figure 10 shows the main screen of the

application with the indication of charging and

battery life management system received from the

battery through the adapter.

Figure 10: SiREV: Charging indication (SOC); EV

estimated autonomy; A charging place identified (Better

Place) at 20 km; and the Context Information (driver,

mode, destination and day period) (in Portuguese

language).

The vehicle's range is 40 km, however the trip

will have a total of 45 km and it will be necessary to

recharge the EV batteries to get to the destination.

The SiREV presents an alert indicating the need

to recharge the batteries, and shows the alternatives

identified by the recommendation system. Since this

is a trip to work, the goal will be to arrive as quickly

as possible, so the recommendation system presents

three alternatives for fast charging, and set the route,

given the previous choices. The first step is to

submit a recommendation to select the candidate

items for the SiREV, and then it uses the

information gleaned from smartgrid to create a list

of all the charging points that can be reached by the

vehicle. In this scenery the vehicle is in Odivelas,

outside Lisbon, so the number of charging points is

high. It is created a list of 30 possible charging sites.

Then the list is sent to the component that manages

the public transport information to validate the

proximity of interfaces with different means of

public transport. The component validates each of

the candidate items and returns the updated list.

The next step is the processing of the

recommendation. At the entrance of the

recommendation system there is a list of 30

candidate items distributed as follows:

• 15 Points of slow charging: 5 of EDP; 5 of

Better Place; 5 of Mobi.E.

• 10 Points of fast charging: 4 of EDP; 3 of Better

Place; 3 of Mobi.E.

• 5 Points of drop-off (in which the batteries are

replaced): 1 of EDP; 3 of Better Place; 1 of

Mobi.E.

The driver of this studied case never chooses the

drop-off sites, nor it is important for him to select

locations near public transport. This driver has a

clear preference for places of charging of EDP,

followed by Better Place, and he never charges his

EV at places of Mobi.E. The algorithm returns a

recommendation to run the utility with more than

three items after comparison with the history of the

driver choices. The following screen is displayed to

the driver, Figure 11:

Figure 11: Showing recommendation, regarding charging

points that fits drivers’ choices (in Portuguese language).

The driver selects the first recommendation of

SiREV and sends the data on the location of the

Public Charging Station (PCS) for the navigation

system. Reached the PCS the charging process starts

and SiREV presents progress and an indication that

the vehicle performs in charging state. It is possible

to visualize in a map the distance that the EV can

reach with that charge. The driver can perform a

charge until the desirable distance is reached (if less

than the maximum range).

After the driver leaves the two children at school,

SiREV continues to put a notification in the

notification area that the vehicle's range is below the

limit. Upon reaching the destination the driver

selects the notice and recommendation system

introduces three new options for charging, this time

the system knows that the vehicle is at its destination

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems

because it is less than 5 km from the site displayed

as a destination, so presents options for charging

along three car parks within walking distance (see

Figure 12-left). The batteries will charge while the

driver is on his job.

After 4 hours, the driver receives an SMS

notification on his phone, indicating that the

batteries are fully charged (with 100% of energy

storage capacity). The driver rests, assured that he

can keep his routine the rest of the day, and that 150

km of EV autonomy are available until the next

working day.

Figure 12: Recommendation of charging places nearby

(left) and SiREV detects that EV range is below 30% of

initial range (right) (in Portuguese language).

6.2 Case 2 - Long Distance between

Home and Work

This scenario presents the situation of a driver who

lives at a considerable distance from his work place.

The driver lives in a distant city but works in

downtown Lisbon. The trip from home to work is

100 km. The driver lives in a building with a garage,

but without a charging point for electric vehicle. A

typical daily schedule of this driver is:

• 08:00 - Departure from home;

• 09:30 - He covers 100 km to his place of work;

• 13:00 - Out to lunch, traveling a total of 5 km;

• 19:30 - Return back home, covering 100 km.

This daily driver travel is 205 km long. Using the

system SiREV, and assuming that, as in the previous

case, in the morning the EV has 25% of its energy

storage capacity. Upon entering the EV and after

selecting the profile and identifying the destination

the driver is alerted to the fact that he must charge

his EV in a radius of 30 km. The system

immediately displays the recommendations taking

into consideration that the driver goes to his work

place (see Figure 12-right).

The three options presented are of fast charging

type, and have in common the fact that they are

service areas that have cafeteria service. Whenever

the driver needs to recharge the EV batteries in the

morning, he chooses a service station with cafeteria

service, because he takes the advantage of taking the

breakfast as the car recharges. While having

breakfast, the driver is alerted to the fact that the EV

batteries are already charged. Back to the car, he

follows his route to employment. Arriving at the

entrance to Lisbon the battery pack is again with a

low level of charge, with only 30% of stored energy,

so that the driver is alerted to that fact.

The diver remains far from his destination, thus

he is presented to all the fast charging options, and

to an option of slow charging, but he is also near an

interface to public transport, enabling the driver to

put the car in a park covered with charging service,

and continuing his journey by public transport. The

driver chooses to deposit the car in the park, let the

EV to be charged, and use public transport to go to

the work.

After 6 hours the driver is notified that the EV

batteries are fully charged, and so, he knows that can

return home with his EV, because the energy stored

in the batteries will be enough to cover the 100 km

back.

6.3 Case 3 - Car Ride at Weekend

In this scenario, the same driver of the second

scenario leaves home on a Saturday morning for a

ride with the family in the western coastal zone of

Lisbon. When entering the vehicle the driver

identifies himself and indicates that he will shift into

drive, with no predetermined destination.

After an hour of travel, the system notifies the

driver that the autonomy of the vehicle reached the

configured limit of 30 km, and displays an alert, as

showed in Figure 13 (left).

In this situation the recommendation system will

consider the following context: (1) Operating mode:

leisure; (2) Time of day: morning; and (3)

Destination: No destination.

The system has the following three charging

station recommendations, see Figure 13 (left).

All recommendations are slow and charging

points have in common the fact that they are near

centers of tourist interest, with places to visit. The

full charge will take about six hours, so the driver

and family can visit the Points of Interest (nearby),

lunch in one of several restaurants. Just four hours

ELECTRICVEHICLEROUTERECOMMENDERSYSTEM

after the driver decides to return home. The vehicle's

range is 70%, and will not reach the destination.

After walking 50 km of new drivers are notified

that you have to carry the load of the vehicle. In this

situation the recommendation system will consider

the following context: (1) Mode of Operation:

leisure; (2) Time of day: Afternoon; and (3)

Destination: House.

By stating that the destination is home, the

recommendation system provides fast charging

points along the route (Figure 13-right), in this

context because the driver always chooses this type

of sites because they want to get home as soon as

possible.

Figure 13: Recommendations given by the Recomender

System, because SiREV detected that range is below the

configured limit of 30 km: On left - recommendation for

slow charging points; On right - recommendation for fast

charging points (in Portuguese language).

7 CONCLUSIONS

The main objective of this work is to help drivers in

the daily use of Electric Vehicles (EV), acting

proactively and increasing the attractiveness of EV.

However, EV technology is still under development,

and so, all available information systems are

necessary for a successful full integration of the

Electric Vehicles in the transportation system. In this

work it is proposed the integration of information

from various areas that can be important in the

successful implementation of a mobility model for

drivers of Electric Vehicles. It is also necessary to

ensure that the integration model is sufficiently

generic and flexible to be adapted to different

systems, having a low impact on the implementation

of a Recommender System. On the other hand, it is

easy to see that the technologies needed to develop a

mobile system, capable of receiving position and

geographic information, and able to establish

connection to external devices or to the Internet,

have a very advanced level of development. The

Android system allows a very simple and efficient

development of applications, with the advantage of

being totally free of cost, both with regard to the

licensing of their own API and developer tools,

which are all based on free software. The ease of

development, the integration with GPS, USB,

Bluetooth, SMS, and the test execution, makes the

implementation work very efficient.

The usage of the developed Recommender

System showed only part of all its potential for

application in the Electric Mobility context, It will

be part of future work to deepen the study of a more

complex model for the Recommender System, with

the optimization of the scenario in question,

demonstrating the validity and enforceability of this

project in a real use, with its added value, in an

attempt to identify functional requirements and

possible ways for their implementation.

ACKNOWLEDGEMENTS

This work is financed by FEDER Funds, through the

Operational Programme for Competitiveness Factors

– COMPETE, and by National Funds through FCT –

Foundation for Science and Technology of Portugal,

under the project PTDC/EEA-EEL/104569/2008 and

the project MIT-PT/EDAM-SMS/0030/2008.

REFERENCES

Ferreira, J. C., Monteiro V. and Afonso J., 2011. Mobile

Geographic Range Prediction for Electric Vehicles.

Proceedings of the 1st Conference in Electronics,

Telecommunications, and Computer Engineering, from

24th to the 25th November, 2011, at Instituto Superior

de Engenharia de Lisboa – ISEL, Lisbon- Portugal.

Gediminas A., (2005). Toward the Next Generation of

Recommender Systems: A Survey of the State-of-the-Art

and Possible Extensions, IEEE Transactions on

Knowledge and Data Engineering, Vol. 17, Nº. 6, June

2005

Tuzhilin A., Gediminas A., 2001. Multidimensional

Recommender Systems: A Data Warehousing

Approach, Springer Berlin/Heidelberg, 2001.

Pereira P., (2010). Sistema de Recomendação para

Condutores de Veículos Eléctricos. MSc dissertation -

ADEETC- ISEL, Lisbon.

Fontes T. F., (2008). Determinação do melhor caminho

em sistemas de transporte terrestre (BPath – Best

Path). Final Year Project at ADEETC – ISEL, Lisbon.

SMARTGREENS2012-1stInternationalConferenceonSmartGridsandGreenITSystems