Professional Documents
Culture Documents
Automotive 5
Automotive 5
ABSTRACT−In this paper, the combined power management/design optimization problem is investigated for a fuel cell/Li-
ion battery PHEV. Formulated as a constrained multi-objective optimization problem (MOP), the combined optimization
problem simultaneously minimizes the vehicle cost and fuel consumption subject to the vehicle performance requirements.
With an emphasis on developing a generic optimization algorithm to find the Pareto front for the synthesized MOP, the Pareto
based multi-objective particle swarm optimization (PMOPSO) algorithm is developed, which solely depends on the concept
of Pareto dominance. Three approaches are introduced to the PMOPSO method to address the constrained MOP. They are: (i)
by incorporating system constraints in the original objective functions, the constrained MOP is transformed to an
unconstrained MOP; (ii) to avoid being trapped in local minima, a disturbance operator with a decaying mutation possibility
is introduced; (iii) to generate a sparsely distributed Pareto front, the concept of crowding distance is utilized to determine the
global guidance for the particles. Finally, under the Matlab/Simulink software environment, simulation results are presented
to demonstrate the effectiveness of the PMOPSO in the derivation of the true Pareto front.
KEY WORDS : Plug-in hybrid electric vehicle, Multi-objective optimization, Particle swarm optimization, Fuel cell,
Component sizing, Power management
645
646 B. GENG, J. K. MILLS and D. SUN
Table 3. Vehicle performance constraint. can be easily converted to –hi'( x) ≤ 0 by putting “−” on both
sides of the inequality.
Constraints Value
Acceleration time 0-60 mph≤12 s 5. PARETO BASED MULTI-OBJECTIVE
Gradeability ≥55 mph@6.8% PARTICLE SWARM OPTIMIZATION
Maximum speed ≥90 mph
Inspired by mimicking social behavior of birds flocking or
All electric range ≥40 miles fish schooling, the basic PSO was initially proposed by
Kennedy and Eberhart (Kennedy and Eberhart, 1995) to
find the optimal solution for unconstrained SOPs. PSO is a
4. COMBINED OPTIMIZATION PROBLEM stochastic population-based approach which uses a swarm
FORMATION of randomly initialized particles to fly through the search
space. At each generation, particles update their velocities
4.1. Vehicle Performance Constraints and positions using knowledge from their past experience
PHEVs must satisfy several general requirements concerning and their neighbors according to (Kim et al., 2010):
their dynamic performance, such as velocity constraint,
acceleration constraint, which are given in Table 3. The “all
electric range” (AER) denotes the maximum distance that (26)
PHEVs travel over a specific driving profile only using the
battery power. In this paper, AER is measured at a constant
vehicle cruise speed of 60 mph (about 100 km/hr) on flat where 1 ≤ n ≤ N is the particle index (N is the total number
roads. of particles in the swarm), k is the generation index. xkn
represents the location of the nth particle. υkn is the velocity
4.2. Constrained MOP Formation vector of the nth particle constrained by υkmin ≤ υkn ≤ υkmax , and
Qbat, Pfc,max and Pmot,max are component sizes that determine υkmax is often set at 10−20% of the dynamic range of the
the vehicle cost and the dynamic performance; Kp and Ki variable on each dimension. Pkn is the personal best for the
are control parameters directly associated with the vehicle nth particle, Gkn is the global best for the nth particle. r1 and
fuel consumption. Rewrite these parameters in the vector r2 are random numbers uniformly distributed in [0,1]. c1 is
form, i.e. x = [Qbat, Pfc,max, Pmot,max, Kp, Ki]T; then x will be the cognitive scaling factor, and c2 is the social scaling
optimized to minimize two objectives as follows: factor. χ is the contraction factor introduced to ensure the
(i) The fuel consumption convenience, and computed by:
N
g1( x ) = ∑ m· H2, k ∆T (22) (27)
k=1
subject to
where, hmax(x) = max[h1(x), h2(x), …, hm(x)], and atan[·] is
Ω:hi( x ) ≤ 0, i = 1, 2, …m (25) the inverse tangent function. hmax(x) is the maximum value
among hi(x). A feasible solution should guarantee that
where Ω gives the feasible domain for the design vector, hi(x) ≤ 0, and hmax(x) ≤ 0 accordingly. However, during the
which is prescribed by m vehicle performance constraints or particle flight, certain particles might fly out of the feasible
design parameter bounds. Interested readers may refer to domain, which results in hi(x) ≥ 0. In that case, hmax(x), by
(Ebbesen et al., 2012a) which gives a detailed explaination on its definition, becomes positive. Then, by minimizing the
how vehicle performance constraints in Table 3 are value of hmax(x) according to Equation (28), hmax(x) will be
transformed to Equation (25). In addition, all performance reduced until hmax(x) ≤ 0; in other words, these particles will
constraints are uniformly expressed in the form of hi(x) ≤ 0 fly back towards the feasible domain. In this way, system
without loss of generality. A constraint in the form of hi'( x) ≥ 0 constraints (25) are considered implicitly by proposing the
650 B. GENG, J. K. MILLS and D. SUN
(29)
set to infinity so that these points may always be selected as Figure 6. Coordinates of the solutions in the external
a global guide, allowing particles to explore wider space. archive.
Step5: Repeat Step 2−4 until kmax or Amax is reached.
Several important parameters for the implementation of
the PMOPSO are listed in Table 5. constrained within their permitted bounds.
Four typical points are selected and compared. a, b and c
6. RESULTS AND DISCUSSIONS are in the Pareto optimal set P, while d is a benchmark
solution the objective of which locates in right region of
To verify the control performance, ten consecutive PF, as shown in Figure 5. The parameters of these four
repetitions of Urban Dynamometer Driving Schedule solutions are listed in Table 6, and their resultant vehicle
(10UDDS) is selected for illustrative purpose. The cost, fuel consumption and dynamic performance criterions
combined optimization problem for the PEMFC PHEV are compared in Table 7. All solutions satisfy the vehicle
over other typical driving cycles can be similarly solved. performance requirements in Table 3. a and b are boundary
Over 10UDDS, the Pareto front (PF) produced by the solutions in P: a results in the smallest vehicle cost and
PMOPSO for the PHEV combined optimization problem is further cost reduction will violate at least one of the
plotted in Figure 5. We can see that by utilizing the concept prescribed vehicle constraints; c leads to the least vehicle
of crowding distance, the non-dominated solutions are
evenly distributed in PF, which implies good solution
Table 6. Parameters of the four selected solutions.
diversity. The Pareto front produced by the PMOPSO
without the mutation operator is also plotted in Figure 5 for Variables Unit a b c d
comparison, which shows that without the mutation Qbat kWh 19.7 22.6 26.8 23.4
operator, the particles are easily trapped in local minima
and cannot converge to the true Pareto front. In addition, Pfc,max kW 62.8 51.2 33.9 47.8
Figure 6 describes coordinates of Pareto optimal solutions Pmot,max kW 115.2 111.8 107.5 111.2
in the archive, showing that all Pareto optimal solutions are
Kp - -427.3 -320.0 -188.2 -507.9
Ki - -1.90 -1.4 -20.7 -432.5
7. CONCLUSION
REFERENCES