You are on page 1of 5

Multi-Objective Optimization of a Parallel Hybrid

Electric Drive Train


Christiane Bertram, Dominik Buecherl, Andreas Thanheiser and Hans-Georg Herzog
Institute of Energy Conversion Technology, Technische Universitaet Muenchen, 80333 Munich, Germany
E-mail: christiane.bertram@tum.de

Since the knowledge of contrary goals’ interdependency is


Abstract — Hybrid electric vehicles are saving raw oil but to very important for an adequate vehicle development a Pareto
achieve this other resources as e.g. copper and lithium are optimization algorithm will be presented and used within this
needed. Therefore the present paper deals with the optimization paper.
of a parallel hybrid electric drive train on both minimal fuel
consumption and minimal use of copper for the electrical
machine and lithium within the electrical energy storage. Since SIMULATION MODEL
copper and lithium are decisive factors during the development
process and fuel consumption depends on the user the Pareto The simulation model used within this paper is a backward-
front will be analyzed looking at different driving cycles. The model of a parallel HEV. The electrical machine is an
chosen algorithm is a hybrid multi-objective optimization induction machine with squirrel cage rotor having a nominal
method of Simulated Annealing, a Genetic Algorithm and power of PEM,basic =17.3 kW and the combustion engine has 6
Tournament Selection. The achieved results of the Pareto cylinders and a nominal power of PICE = 225 kW. Further, the
optimized HEV drive train are presented and the
energy storage is a lithium ion battery. The parameters of the
interdependency of those goals is analyzed.
optimization are an axial and a radial scaling factor (ka and kr)
of the electrical machine, the energy content of the lithium
INTRODUCTION ion battery and the factor of load point increase of the
decision based control strategy. The objectives of the
In many cases the fuel consumption of hybrid electric optimization are fuel consumption, lithium usage within the
vehicles (HEVs) was optimized by either optimization of the energy storage and copper usage within the electrical
control strategy or of the drive train components [1, 2]. Since machine. The drive train is illustrated in Figure 1.
the reduction of fuel consumption and emissions is the mayor
advantage of HEVs a multi-objective optimization on The copper needed (mCopper) for the electrical machine
different emission types and fuel consumption was done in depends on the power of the electrical machine as it may be
[3-5]. But there are more components needed in case of a seen in equation (1). Table I gives relevant values of the
HEV than in case of a conventional car and thereby other raw simulation model’s parameters. A detailed description of the
materials, as copper or lithium, are consumed. In order to electrical machine’s scalable model is given in [15].
reduce the consumption of all raw materials contrary goals
have to be pursued. Looking at a conventional car and a HEV  = ;
∙  ∙ 
it is obvious that the conventional car consumes more fuel  = 2 ⋅ 
 +   ⋅ 24 + 36 ⋅  
and the HEV on the other hand needs more Lithium (in case +   1!
of a Lithium-ion battery). As a consequence multi-objective
optimization methods are needed and the interdependency of TABLE I
the contrary goals has to be analyzed [5]. RELEVANT PARAMETERS OF THE SIMULATION MODEL
Parameter Description Value
Within this paper an optimization algorithm capable of ka Axial scaling factor 0.75 – 1.5
those optimization problems looking for the whole Pareto kr Radial scaling factor 0.75-1.1
front will be presented and drawn out on the optimization " # kg
Density of copper 8.96 ⋅ 10%& + &
problem of a parallel HEV which has to be optimized on cm
mring Copper mass of the squirrel 3 &
93.3 cm ⋅ k , ⋅ ρ./
minimal fuel consumption as well as on minimal copper and
cage’s end ring
lithium usage looking at diverse driving cycles.
mrcon Copper mass of rotor 16.02 cm& ⋅ k 1 k , ρ./
conductor
The aim of the paper is for one thing to show the necessity of mscon Copper mass of stator 13.48 cm3⋅ k 1 ρ./
multi-objective drive train optimization, for another thing to conductor
explain an algorithm capable of those optimization problems mwcon Copper mass of winding head 2.01 kg
finding the Pareto front. In many cases multi-objective conductor
optimization problems are transformed to single objective mLiCell Lithium of one cell 52 g
optimization problems due the usage of weighted sums [5, 6]. nCells Number of cells 10-100

978-1-61284-247-9/11/$26.00 ©2011 IEEE


A. Optimization process
The algorithm used within this paper is a hybrid
optimization method of a Genetic Algorithm (GA) and
Simulated Annealing (SA). Genetic Algorithms emulate
biological evolution and Simulated Annealing mimics the
annealing process of metal. GA and SA are often combined
during optimization since the SA may prevent the premature
convergence of the GA [8-12]. The SA is thereby used during
the selection-phase of the GA.

In order to find the Pareto front the principle of tournament


selection was chosen, thereby two individuals of the
population are chosen and compete against each other. If
Figure 1: Topology of the simulated drive train individual 1 is dominated by individual 2, meaning that
individual 2 is in at least one of the objectives better than
The lithium needed depends on the number of lithium ion individual 1 and in none worse, individual 2 is chosen and the
cells and thereby on both the energy content and the voltage other way around [13]. If neither of the individuals is
level of the energy storage. The energy content of one cell dominated the selection uses Simulated Annealing in order to
amounts to 162 Wh. According to [16] the Lithium needed reduce the risk of preliminary convergence. Figure 3 shows
amounts to 320 g/kWh i.e. about 52 g of Lithium is needed for the flowchart of the optimization algorithm.
each cell. The lithium needed is calculated by equation (2).
During the Thermodynamic Tournament Selection equation
2
34
#5 = 2
66 766 2! (3) is used as it was done in [8] in cases where both or non
The additional mass of all hybrid components are taken into individual is dominated.
account. Also the additional mass due to the scaling of the

electrical machine and the scaling of the energy storage are ∑
%A C
6DA :6 + :4
taken into account during the calculations. The total mass of 8 = 〈:〉 − = > ?@A = − = > ?@A E, ℎ! 3!
the conventional car amounts to 1170 kg. The simulation E
@ @DA
model is optimized by the hybrid multi-objective Where F is the current individual’s free energy, Ei is the ith
optimization method of Simulated Annealing, a Genetic individual’s energy and T is the current temperature. The
entropy ?@A is calculated by equation (4) where H@ is the
Algorithm and Tournament Selection which will be explained
in detail in the following section.
probability of j on the locus k [8].

OPTIMIZATION ALGORITHM
?@A E, ℎ! = > H@ E, ℎ! log H@ E, ℎ! 4!
The main challenge in case of multi-objective optimization H ∈{M,A}
is that the Pareto optimal results are equivalently good but
different. A solution is called Pareto optimal if the
improvement of one of the objectives leads to a worsening of B. Validation of the Algorithm
at least one of the other objectives, a two dimensional In order to validate the functionality of the algorithm it was
example of such a Pareto front is given in Figure 2 [5]. Due to drawn out on a test function. Figure 3 shows the simulation
this the results of a Pareto optimization may not be sorted as result while optimizing the test function given in equation (5)
it is done usually. Therefore methods to select the “better” [14].
result during the optimization process as the non-dominated
sorting or tournament selection have been suggested [7]. OA PA , P ! = −−PA + P !
1
O PA , P ! = − Q PA + P + 1R
2
0 ≤ PA ≤ 6, 0 ≤ P ≤ 6. 5!

The Pareto optimal solutions of this test problem are


0 ≤ PA ≤ 6, P = 6. The result of the parameter values is
illustrated in Figure 4. Hence the algorithm was proven to be
capable of multi-objective optimization problems it was
carried out on the drive train simulation. The results of the
optimization of fuel consumption, copper mass and needed
lithium are given in the following section.
Figure 2: Pareto optimal front [5]
Figure 4: Identified optimal parameter values in case of the
test function

DRIVE TRAIN ANALYZES


Considering a conventional car and the same as a pure
electric vehicle having a lithium battery both cases are Pareto
optimal if the objectives of power train optimization are fuel
consumption and lithium usage. Both cases are the most
extreme results and the challenge for the engineer lies in
finding compromises being part of the Pareto front satisfying
the additional requirements best. Since the driving cycle has a
huge impact on the optimization of HEVs the optimization
was done for the NEDC, the FTP-75 and the ARTEMIS
driving cycle in the following.

A. Results in Case of NEDC


Figure 5 illustrates the found Pareto front of the objectives
fuel consumption, additional copper usage (resulting from the
copper needed in the electrical machine) and lithium needed
for the electrical energy storage while simulating the NEDC.
The red diamond illustrates the fuel consumption of the
conventional equivalent vehicle. Since the electrical machine
may be scaled only in a certain range the Pareto front
illustrated in Figure 5 as blue stars does not regard the
conventional version of the car. The conventional car’s result
is only given as a reference point. From the results illustrated
it becomes quite clear that the fuel consumption increases a
lot in case of a small electrical energy storage where less than
0.5 kg lithium are needed but also stays more or less constant
when the energy storage becomes larger. In those cases of an
extremely small energy storage the fuel consumption even
exceeds the consumption of the conventional car which is due
to the fact that the limited voltage level of such an energy
storage makes the hybrid functions useless and the additional
mass of the electrical machine leads to this increase of fuel
consumption.

Looking at the results it becomes further visible that the


Figure 3: Flowchart of the multi-objective TDGA combined fuel consumption only differs in most cases in a small range
with Tournament selection since the minimal fuel consumption achieved amounts to
4.8 l/km while the median is 5.9 l/km and the maximal fuel
consumption amounts to 14.9 l/km. This also shows the
interdependency of the machines power and the voltage level
of the energy storage. If the machine needs a higher voltage C. Results in Case of ARTEMIS
level than the energy storage may provide in most cases the Figure 7 illustrates the optimization results achieved during the
hybrid functions may not perform adequately. simulation using the ARTEMIS driving cycle. As before the red
diamond once again shows the fuel consumption of the conventional
equivalent vehicle but was not considered during optimization.

Figure 5: Optimization results in case of NEDC

Since the NEDC is a synthetic driving cycle further optimization


was done using the FTP-75 and the results are shown in the
following. Figure 7: Optimization results in case of ARTEMIS

B. Results in Case of FTP-75 In case of the ARTEMIS cycle the values of the fuel
Figure 6 illustrates the optimization results achieved during the consumption are even closer than in the other two cases. Here the
simulation using the FTP-75. As in Figure 5 the red diamond once minimal fuel consumption amounts to 7 l/km while the median is
again shows the fuel consumption of the conventional equivalent 7.2 l/km and the maximum lies at 12 l/km. Also here the
vehicle but was not considered during optimization. The results are interdependency between energy storage and machine size is visible.
similar to the results in case of the NEDC. Due to the higher power
demand of the driving cycle the optimal copper mass reaches higher It is also visible that the lithium needed and the copper mass are
values since the power of the machine and the copper mass depend lower in this case than looking at the results of the NEDC and FTP-
on each other as it was shown in Section 2. 75. This is because the ARTEMIS driving cycle has such a huge
power demand that the machine used within this paper could also in
case of maximal scaling satisfy the power demand of the driving
cycle. Therefore both the electrical machine and the electrical energy
storage are scaled down during the optimization process.

CONCLUSION
Within this paper it was shown how important multi-
objective optimization of hybrid electric vehicles is. The
multi-objective optimization was drawn out on the model of a
parallel HEV looking at the minimal copper needed, lithium
usage and fuel consumption while simulating the NEDC,
FTP-75 and the ARTEMIS driving cycle.

Since several Pareto optimal solutions have been found during


the optimization process optimal solutions may be selected from the
Pareto front with respect to the driver’s needs. In cases of high
Figure 6: Optimization results in case of FTP-75 power demand e.g. in the mountains or in cases where the driver
often has to go on the interstate the power of the electrical machine
The ARTEMIS driving cycle has an even higher power should be higher as the results of the FTP-75 showed. In cases of
demand than the FTP-75 therefore the analyses done using city cars the results show that a small electric machine of 17 kW or
the ARTEMIS driving cycle are shown in the following less may reduce the fuel consumption significantly as it was shown
subsection. in case of NEDC.
Especially the results of the ARTEMIS driving cycle Transactions on Evolutionary Computation, vol. 14, no. 4, pp. 580-590,
showed that the power of the electrical machine used within Aug 2010.
this paper is too low even in case of maximal scaling for [11] T. Renyuan, Y. Shiyou, L. Yan, W. Geng, and M. Tiemin, “ Combined
some analyses. Strategy of Improved Simulated Annealing and Genetic Algorithm for
Inverse Problem”, in IEEE Transactions on Magnetics, vol. 32, no. 3,
Therefore in future work the optimization process should pp. 1326-1329, May 1996.
be carried out using higher machine powers. Also further
[12] H. Kita, Y. Yabumoto, N. Mori, and Y. Nishikawa, “Multi-Objective
parameters as e.g. the weight of the car should be varied and
Optimization by Means of the Thermodynamical Genetic Algorithm”,
the influence on the Pareto Front of such changes should be
Parallel Problem Solving from Nature — PPSN IV Lecture Notes in
analyzed. Additionally detailed usage scenarios should be
Computer Science, 1996, vol. 1141/1996, pp. 504-512, DOI: 10.1007/3-
built up to show the pros of searching the Pareto front.
540-61723-X_1014
[13] Y. Wang, Z. Cai, G. Guo, and Y. Zhou, “Multiobjective Optimization
and Hybrid Evolutionary Algorithm to Solve Constrained Optimization
REFERENCES Problems”, in IEEE Transactions on Systems, Man, and Cybernetics —
Part B: Cybernetics, vol. 37, no. 3, June 2007.
[1] D. W. Gao, C. Mi, and A. Emadi, “Modeling and Simulation of Electric
[14] K. Deb, L. Thiele, M. Laumanns, and E. Zitzler, “Scalable Multi-
and Hybrid Vehicles,” in Proceedings of the IEEE, vol. 95, no. 4,
Objective Optimization Test Problems”, in proceedings of the 2002
pp.729-745, 2007.
IEEE Congress on Evolutionary Computation, pp. 825-830, May 2002.
[2] C. Bertram, D. Buecherl, T. Kohler, and H.-G. Herzog, “DIRECT as
[15] D. Buecherl and H.-G. Herzog, “Iron Loss Modeling by Complex
Two-Level Optimization Method for Drive Train Design and Control of
Inductances for Steady State Simulation of Electrical
Hybrid Electric Vehicles” in Proceedings of the EVS, 25nd
Machines,”Proceedings of the 20th International Symposium on Power
International Battery, Hybrid and Fuel Cell Electric Vehicle Symposium
Electronics, Electrical Drives, Automation and Motion, 883, Jun. 14-16,
& Exhibition, Nov. 2010.
2010.
[3] L.-C. Fang and S.-Y. Qin, “Concurrent Optimization for Parameters of
[16] W. Tahil, “How Much Lithium does a LiIon EV battery really need?,”
Powertrain and Control System of Hybrid Electric Vehicle Based on
Meridian International Research: http://www.meridian-int-
Multi-Objective Genetic Algorithms”, in proceedings of the IEEE
res.com/Projects/How_Much_Lithium_Per_Battery.pdf Downloaded:
SICE-ICASE, pp. 2424-2429, Oct. 2006.
2011/06/29
[4] B. Huang, Z. Wang, and Y. Xu, “Multi-Objective Genetic Algorithm for
Hybrid Electric Vehicle Parameter Optimization”, in proceedings of the
IEEE International Conference on Intelligent Robots and Systems, pp.
5177-5182, Oct. 2006.
[5] C. Desai and S.S. Williamson, “Optimal Design of a Parallel Hybrid
Electric Vehicle using Multi-Objective Genetic Algorithms”, in
proceedings of the 5th IEEE Vehicle Power and Propulsion Conference,
pp. 871-876, Sept 2009.
[6] T. Hofman, D. Hoekstra, R.M. van Druten, and M. Steinbuch, “Optimal
design of energy storage systems for hybrid vehicle drivetrains”, in
proceedings of the 1st IEEE Vehicle Power and Propulsion Conference,
pp. 73-77, Sept. 2005.
[7] J. Horn, N. Nafplioris, and D.E. Goldberg, “A Niched Pareto Genetic
Algorithm for Multiobjective Optimization”, in proceedings of the 1st
IEEE Conference on Evolutionary Computation, pp. 82-87, June 1994.
[8] N. Mori, J. Yoshida, H. Tamaki, H. Kita, and Y. Nishikawa, “A
Thermodynamical Selection Rule for the Genetic Algorithm”, in
proceedings of the 2nd IEEE Conference on Evolutionary Computation,
pp. 188-192, Nov. 1995.
[9] S.-J. Lin, Y.-J. Huang, and C.-M. Ko, “MCM Placement Problem with
GASA Multi-objective Optimization Strategy”, in proceedings of the
IEEE International Conference on Electronic Materials and Packaging,
Nov. 2007.
[10] R. Bai, E. K. Burke, G. Kendall, J. Li, and B. McCollum, “A Hybrid
Evolutionary Approach to the Nurse Rostering Problem”, in IEEE

You might also like