Copyright © IFAC Intelligent Components for Autonomous and
`Semi-Autonomous Vehicles. Toulouse. France. 1995
`
`PREDICTING THE USE OF A HYBRID ELECTRIC
`VEIDCLE
`
`C P Quigley, R J Ball, A M Vinsome.
`
`Dr. RP Jones.
`
`Warwick Manufacturing Group,
`Advanced Technology Centre,
`University of Warwick.
`Coventry CV4 7AL. u.K..
`Tel: +44(0)1203523794
`Fax: +44(0)1203523387
`E-Mail: c.p.quigley@atcmail.warwick.ac.uk
`
`Department of Engineering.
`University of Warwick,
`Coventry CV4 7AL, u.K..
`Tel: +44(0)1203523108
`Fax: +44(0)1203418922
`E-Mail: pj@eng.warwick.ac.uk
`
`Abstract: This paper outlines the initial stages of a project to analyze the requirements and
`then design an intelligent controller for hybrid electric vehicles. Such a controller would be
`required to manage energy flow through the hybrid drive train and for optimum control
`would require a number of parameters normally available only upon journey completion.
`This paper presents work to attempt to predict these parameters at the start of the journey
`using intelligent classification techniques and a knowledge base of previous journey
`histories.
`
`Keywords:
`
`Hybrid Vehicles, Prediction Methods, Automotive Control, Intelligent
`Control, Navigation Systems.
`
`1. INTRODUCTION
`
`Much of the research into new forms of vehicle
`propulsion is motivated by legislation intended to
`limit
`the polluting effects of vehicle exhaust
`emissions. Future legislation in both the USA and
`European Community will introduce progressively
`more stringent limits on vehicle exhaust emissions
`over the next 15 to 20 years. Ultimately, there will be
`a requirement for vehicle manufacturers to supply
`Zero Emission Vehicles (ZEVs) and Low Emission
`Vehicles (LEVs) for use as a form of private
`transport within large urban conurbations. At present
`electric vehicles are the only practical candidates as
`ZEVs, whilst hybrid electric vehicles currently form
`the most serious contenders as LEVs.
`
`Hybrid electric vehicles have a propulsion system
`which includes a heat engine, and one or more
`electric motors and/or generators with an associated
`
`traction battery. The propulsion system in a hybrid
`electric vehicle can be assembled in a variety of
`configurations. One possible arrangement is a parallel
`hybrid vehicle consisting of a single electric motor
`(E) and heat engine (HE) mechanically coupled to a
`single drive shaft. A typical parallel hybrid vehicle
`configuration is described in figure 1. Power from the
`electric motor and/or heat engine is transmitted to the
`road wheels by drive shafts and gear mechanisms.
`Power for the electric motor is supplied by the
`traction battery (B). If the motor is driven either
`directly from the heat engine or during braking, it will
`generate current to charge the traction battery.
`
`In its simplest form, the power from the two drives
`can be provided in one of three modes:-
`
`I) Electric motor only.
`2) Heat engine only.
`3) Electric motor and heat engine combination.
`
`129
`
`BMW1054
`Page 1 of 6
`
`

`

`For example in the UK, 16% of commuters in the
`London area drive their cars to work, whereas outside
`London almost 50% of commuters drive to work
`(National Travel Survey, 1993).
`
`2. HYBRID POWER TRAIN CONTROL
`
`As previously stated, the proposed controller will
`allow journey parameters to be reliably estimated
`upon journey departure. In order to do this the
`information available to the controller could take one
`of two forms:-
`
`1 st Generation Control
`The essence of this type of control is
`that all
`information would only be available internally to the
`vehicle, from transducers belonging to the vehicle. A
`controller of this type, if implemented, would use
`signals derived from technology already present in
`modem day vehicles (e.g. electronic tachometer,
`engine management system).
`
`Such information would include:-
`
`a) Drivers Operational Inputs:-
`
`b) Time of day/year.
`c) Engine Management Data:-
`d) Road speed.
`
`Throttle
`Brake etc.
`
`Engine speed etc.
`
`2nd Generation Control
`This type of control would employ the use of 1st
`generation control information, but also would have
`the
`additional
`advantage of vehicle
`location
`information relayed into the vehicle from external
`sources. Such external sources could take the form of
`a GPS
`(Global Positioning System) navigation
`system, or a road transport telematic infrastructure of
`the future, perhaps employing the use of road side
`beacons. Much research is currently underway into
`the use of such systems for road transport. GPS
`systems have been suggested for use in vehicles for a
`variety of applications; for example rapid vehicle
`location in the event of a road accident (Voger and
`Harrer, 1994). Road side beacons have been a
`suggested tool not only for automatic debiting in road
`tOlling schemes, but also
`for
`interactive route
`guidance systems (Bueno and Ongaro, 1991). A 2nd
`generation controller would explicitly know
`its
`location by receiving vehicle location information via
`such systems.
`
`Information available:-
`a)All 1st generation control information.
`b)Vehicle
`location
`information
`i.e.
`latitude and
`longitude.
`
`Control Decisions.
`An intelligent hybrid electric powertrain controller
`would have to make a decision of control strategy in
`
`figure 1. Structure of a typical parallel hybrid vehicle.
`
`Using the electric motor or heat engine exclusively
`(modes 1 and 2) present a manageable control
`problem for the driver of the vehicle, but their
`combined use (mode 3) makes it very difficult for the
`driver to control optimally. Previous work at the
`University of Warwick (Farrall and Jones, 1993;
`Farrall, 1993) has investigated the use of fuzzy
`decision making for the management of energy flow
`within a hybrid electric vehicle in this third mode. It
`was concluded that fuzzy control could provide
`benefits over a limited range of operation, but in
`order to obtain better performance over the complete
`range of operation, a method of adapting the fuzzy
`rules would be required.
`
`To enable a hybrid electric power train controller to
`adapt to a wide variety of vehicle operation many
`parameters not normally used in vehicle control
`systems would be required , e.g. Journey duration,
`journey distance,
`time of departure,
`journey
`destination. Unfortunately most of these parameters
`are only known upon completion of a given journey.
`Therefore a means of intelligently estimating these
`parameters, based on the controller's past experience
`is needed.
`
`A programme still very much in its infancy is the
`design of an intelligent controller. The proposed
`controller will allow journey parameters to be reliably
`estimated upon journey departure, and
`therefore
`allow for optimal operation with respect to exhaust
`emissions and fuel consumption.
`
`The successful implementation of such a controller
`relies on the idea that many cars will have habitual
`usage characteristics for a high percentage of their
`journeys, and hence
`the ability
`to predict
`the
`occurrence of a
`journey and
`its
`associated
`characteristics will be quite high. A commuter
`journey is a particularly good example of a journey
`that exhibits habitual characteristics, and in the UK
`accounts for around 20% of all journeys taken by car
`(National Travel Survey, 1993). Other journey types
`(Business, Education, Shopping, Leisure) may also
`exhibit habitual characteristics, therefore the benefits
`to be gained from the use of an intelligent controller
`are high. The impact of such benefits will depend
`upon the geographical location the vehicle is used in.
`
`130
`
`BMW1054
`Page 2 of 6
`
`

`

`two situations for the hybrid electric vehicle; Where
`Am I Going? and Where Am I Now? The prediction
`of vehicle usage is concerned only with the Where
`Am I Going? decision. Where Am I Now? is a means
`by which an intelligent controller would continually
`reassess its original prediction.
`
`To the knowledge of the authors, the work described
`here
`involving
`the
`prediction
`of
`journey
`characteristics upon departure is unique.
`
`3. VEHICLE EXPERIMENTS
`
`Where Am I Going? is the initial estimation on the
`type of journey at the time of departure. The initial
`decision here is based on the Expectation of a journey
`type. Infonnation available at
`the
`time of this
`decision differs between 1 st and 2nd generation
`control.
`
`for
`the project
`throughout
`required
`is
`Data
`investigation into methods of vehicle use prediction.
`This is achieved by the use of a data logger based
`around a GPS navigation system. Data recorded by
`the logger can be considered in terms of 1st or 2nd
`generation control data as follows:-
`
`1st Generation Control:- This can only use the
`present system time (time of day, day of week), the
`only infonnation that is available at this instant. A
`journey is expected only if it frequently occurs at the
`same time of day, e.g. a morning commuting type
`journey.
`
`2nd Generation Control :- This has the additional
`infonnation of the vehicle' s ground position at the
`time of departure. Expectation here is based on time
`of day, day of week and the vehicle's present global
`position.
`
`If the controller decides a journey is expected it can
`make an estimation of
`the expected
`journey
`parameters, and an appropriate optimized control
`strategy can be referenced from the controller's
`memory. If a journey is not expected, the controller
`will choose the use of a general purpose control
`strategy,
`thus
`providing
`reasonably
`efficient
`operation only. Figure 2. shows the decision process.
`
`Jeffcoate, 1971 ;
`(Smeed and
`Previous studies
`Hennan and Lam, 1974) have examined commuter
`journeys in different geographical locations. They
`have found that they can describe the travel time
`variability on
`these routes mathematically with
`variables such as departure time and journey distance.
`
`1 st Generation Control Parameters
`Time of departure,
`Journey time elapsed,
`Speed over ground, derived from latitude and
`longitude.
`
`Additional Parameters for 2nd Generation Control
`Explicit vehicle location (i.e. Latitude, Longitude)
`Bearing relative to north, derived from latitude and
`longitude.
`
`The GPS data logger is described in Figure 4. The
`GPS system requires signals from 3 satellites in order
`to obtain a 2-D position fix, and at least 4 satellites to
`obtain a 3-D position fix.
`
`Careful consideration has gone into the selection of
`subjects. They are being chosen so that the main
`vehicle user represents a subset of the UK driving
`population, and
`their selection is based on age and
`sex statistics from UK driving licence registrations
`(National Travel Survey, 1993) and occupational
`statistics (Labour Force Survey Quarterly Bulletin,
`1994). The subjects are being selected to give a
`spread
`of
`different
`occupations,
`different
`geographical location and different hours of work
`(shift, fixed hours, flexible time, part-time etc.). As a
`result, twenty subjects will have the GPS data logger
`installed in their car for a period of one month each.
`
`Vehicle Ignition On
`
`1st Generation
`Get System Time
`
`2nd Generation
`Get Vehicles Present Global
`Position
`
`Cigarette Lighter I
`[
`Socket +12v
`
`Serial Port
`
`General
`Purpose
`Control
`Algorithm
`
`GPS Navigation
`System
`
`Power Supply
`
`I
`
`Laptop Personal
`Computer
`
`r on parcel she
`
`External Ante nna,
`nted
`usually mou
`If at
`rear of vehicle.
`
`Figure 2. Decision Process of Where Am I Going?
`
`Figure 4. GPS Data Logger set-up
`
`131
`
`BMW1054
`Page 3 of 6
`
`

`

`4. PRELIMINARY FINDINGS
`
`At the time of writing, only the results from one
`vehicle were available. All of the journeys taken in
`this vehicle have been logged over a one month
`period; a total of 125 journeys. Figure 5 shows an
`example of the form of the raw data for latitude and
`longitude obtained on a single journey. Speed over
`ground and bearing are derived from this data.
`
`Time Elapsed (seconds)
`
`Arrtvol ---/"v-'-
`
`/
`
`~
`/
`/
`
`I
`'
`\
`~/
`
`Ci) 4.0
`.l!!
`::l 1.
`e:: ·E 3.D
`-8 1.
`
`::l
`:t::: 2.0
`
`1.0
`
`Departure
`
`/
`
`'-/'[
`
`\
`
`Cl e:: o 1.$
`...J
`.S:
`Q)
`Cl D.'
`e::
`~ 0.0 L..~-_'--~-"''----''''':---:'8OD'':-~,-:'.'DDD=---'::':'''''::--
`U
`D
`~
`Time Elapsed (seconds)
`Figure. 5 Raw Data Obtained from GPS Data Logger.
`
`All Days
`
`Weekend
`Days
`Only
`
`Week
`Days
`Only
`
`6
`
`- .
`
`.·1·· ·
`
`i -
`
`~ :>
`0 ...
`
`500°
`C(f~~l ·
`'000
`10"
`f ...
`<'0,,0' .. )
`
`'500
`
`2500
`
`-:.
`.. .. ~ .
`
`L , ,
`
`.. , ,
`
`. . - .. " ' .
`
`A summary of initial data analysis is presented (i.e.
`journey duration,
`journey distance, and ground
`locations visited). The data is considered to develop
`rules to assist in journey prediction. 1st generation
`control data is considered first. This is followed by
`considering any additional advantages gained by
`using 2nd generation control.
`
`1st Generation Control Data
`As it is the case that many people work a five day
`week Monday to Friday, the data has been split into
`two distinct subsets for consideration; weekdays and
`weekend. Figure 6 shows the duration of our subjects
`journeys over the month period.
`
`The weekend distribution shows a no obvious pattern,
`whereas the week day plot shows a number of
`journeys occurring between the hours 07 .00 - 08 .00,
`with a duration between 1000 - 1300 seconds. This
`actually corresponds with
`the subjects morning
`journey to work.
`
`Figure 6. Variability of Journey Duration
`
`Figure 7 shows the distribution of the distance
`covered on the vehicle' s journeys. Again the weekend
`distribution shows no obvious pattern. The week day
`plot shows a number of journeys occurring between
`the hours 07.00 - 08.00, with a distance around 14km.
`This correlates well with the data obtained for
`journey duration. The journeys occurring between
`07.00 - 08.00 hours in fact account for only 13.6% of
`the journeys in one month.
`
`From this data we can deduce the following rule on
`journey expectation:-
`
`132
`
`BMW1054
`Page 4 of 6
`
`

`

`If it is a weekday, and the time is between 07.00-
`OB.OOa.m.
`then
`
`there is a high expectation of a journey of
`1000 to 1300 seconds duration, with a distance
`around 14km.
`
`!!
`c:
`" 8
`
`•
`
`3
`
`J
`All Days
`
`Weekend
`Days Only
`
`.' "
`
`. j
`
`:-1 .
`
`:. r
`
`-' . ~ . ~ ,
`-"'\ -
`
`~
`
`, . , -
`
`Week
`Days
`Only
`
`4
`
`!!
`c:
`" o
`
`u
`
`s
`
`Figure 7. Variability of Journey Distance
`
`2nd Generation Control Data
`Plotting the cumulative distribution of vehicle global
`position (expressed in latitude and longitude) both at
`journey departure and arrival, gives an indication of
`the locations regularly visited. Figure 8 shows the
`distributions for all days, weekend days only, and
`week days only.
`
`The weekend distribution shows a number of
`locations visited covering a large geographical area.
`Only two of these locations are regularly visited
`(Positions A and B). The week day plot shows only
`locations (Positions A, B and C).
`three visited
`
`Although this information is useful, the distributions
`in figure 8 do not show how the locations are related.
`
`All
`
`. , _ . . -
`
`·
`
`i
`• .
`
`i
`
`1~
`
`120
`
`lOO
`
`eo . c
`2~0 _~. La_
`
`"
`0 u
`
`80
`
`~
`
`20
`
`15
`
`lOO
`
`30
`
`( - I
`
`• i
`
`. '" .
`
`Days
`Only
`
`20
`
`40
`
`(~",
`
`.)
`
`Week
`Days
`Only
`
`",
`
`30
`
`lOO
`
`, ,
`
`le
`li
`.
`
`, ,
`",
`
`!
`
`~
`
`&0
`
`60
`
`c
`
`" ~ 8
`
`20
`
`o 0
`
`eo
`
`60
`
`. .. ~ 8
`
`20
`
`Figure. 8 Area Covered by Vehicle During Logging
`Period
`
`Nodal Analysis (Figure 9) gives a better view of how
`each location is inter-related. In the example shown,
`locations A, Band C account for about 62% of the
`journey destinations for this subject. Optimization
`from journey prediction could provide quite high
`benefits if the interconnecting journeys to
`these
`locations could be predicted. Each node has latitude
`and longitude as its properties (not shown on the
`diagram), each arc of the nodal diagram has the
`following properties associated with it:-
`
`Length of journey.
`Duration of journey.
`Usual times of departure, e.g.
`Day of week
`Time of day
`
`133
`
`BMW1054
`Page 5 of 6
`
`

`

`period. Nodal analysis of the locations visited by the
`vehicle, will be very useful if a 2nd generation
`controller is to be implemented (i.e. in terms of the
`number of nodes to be accommodated for), but
`ultimately we may not have the lUXury of such a
`system. Initial results suggest that implementation of
`a
`I st generation controller will be
`the most
`challenging, but this could be the least rewarding in
`terms of the number journeys that are predictable for
`a given vehicle. Further work will
`include a
`continuation of the vehicle logging programme and a
`more in depth examination of the data.
`
`ACKNOWLEDGMENTS
`
`The authors acknowledge the support provided by
`Rover Group for this work. The work was funded in
`part by
`the UK Engineering Physical Sciences
`Research Council (Grant No. GR/K35976).
`
`REFERENCES
`
`Bueno, S. and Ongaro, D. (1991), "VehiclelRoadside
`Communication
`for Route Guidance",
`Proceedings of the DRIVE Conference,
`Brussels, VoI.l, pp4-6.
`Farrall, S. D. and Jones, R. P. (1993), "Energy
`management in an automotive electriclheat
`engine hybrid powertrain using
`fuzzy
`decision making", Proceedings of 1993
`IEEE
`International
`Symposium
`on
`Intelligent Control, pp463 - 468.
`Farrall, S. D. (1993), "A Study in the Use of Fuzzy
`Logic in the Management of an Automotive
`Heat EnginelElectric Hybrid Vehicle
`Powertrain", Thesis (Ph.D.) - University of
`Warwick.
`Herman R, Lam T (1974), "Trip Time Characteristics
`of Journeys To
`and
`From Work,
`Proceedings
`of the Sixth International
`Symposium on Transportation and Traffic
`Theory, pp57-85.
`Labour Force Survey Quarterly Bulletin (Dec. 1994),
`The Government Statistical Service No.lO,
`pp6.
`National Travel Survey 1989/91, Transport Statistics
`Report, HMSO, London, September 1993.
`Smeed, R. J. and Jeffcoate, G. O. (1971), "The
`Variability of Car Journey Times on a
`Particular Route", Traffic Engineering and
`Control, Vol.13, pp238-243.
`Vogal D, Harrer S
`(1994), "DGPS - Emergency
`Location System for Vehicles", The Journal
`of Navigation, Vo1.47, pp349-360.
`
`'Other Location' represents various other locations
`that have been visited during the logging period by
`the vehicle, but are not regularly visited, therefore
`appear unpredictable at this stage. 'Other Location',
`if unpredictable, would be a 'Don't Care' situation
`and would result in the use of a default control
`strategy in the final controller.
`
`~I
`
`"The numbers at each arc represent the number of journey occurrences.
`
`Figure 9 Nodal Analysis of Locations Regularly
`Visited for All Days in One Month Logging Period
`
`Using all the data collected (i.e. both 1st and 2nd
`Generation control) the following rule can be
`deduced
`
`If it is a weekday, and the time is between 07.00-
`08.00 a.m. and I am at Position A
`then
`
`there is a high expectation of a journey to
`Position C for a duration of 1000 to 1300 seconds
`over a distance around 14km.
`
`From the limited data collected, we cannot yet state
`reliably how much higher the expectation of a
`particular journey would be using 2nd generation
`control.
`
`5. CONCLUSION
`
`It is seen from the initial results from only the first
`subject, that it is possible to construct simple rules
`that could form the basis of an intelligent controller.
`This motivates the need for further research. Nodal
`analysis has shown that the first subject to take part in
`vehicle experiments predominantly visits just three
`locations, accounting for more than half of the
`journeys taken during the one month data logging
`
`134
`
`BMW1054
`Page 6 of 6
`
`