Вы находитесь на странице: 1из 9

Applied Energy 165 (2016) 676684

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

A comparative study of biodiesel engine performance optimization using


enhanced hybrid PSOGA and basic GA
Qiang Zhang a, Ryan M. Ogren b, Song-Charng Kong b,
a
b

School of Energy and Power Engineering, Jiangsu University of Science and Technology, Zhenjiang, China
Department of Mechanical Engineering, Iowa State University, Ames, IA, USA

h i g h l i g h t s
 Performance optimization of a CI engine fueled with soy biodiesel was conducted.
 Both GA and hybrid PSOGA algorithms were tested.
 The present hybrid PSOGA algorithm was proven an effective tool for engine combustion optimization.
 The dynamometer time, fuel economy, and exhaust emissions were improved by use of the hybrid PSOGA algorithm.

a r t i c l e

i n f o

Article history:
Received 29 May 2015
Received in revised form 18 November 2015
Accepted 12 December 2015

Keywords:
Biodiesel engine performance
Particle swarm optimization
Genetic algorithm

a b s t r a c t
Efficient optimization algorithms are critical to the development of new engine technology. In this study,
experimental investigations were carried out on optimizing the performance of a four-cylinder,
turbocharged, direct-injection diesel engine running with soy biodiesel. An effective hybrid particle
swarm optimization (PSO) and genetic algorithm (GA) method using a small population was developed
and tested to optimize five operating parameters, including EGR rate, pilot timing, pilot ratio, main injection timing, and injection pressure. Based on the measured engine performance and emissions, results
show that the new hybrid algorithm can significantly speed up the optimization process and achieve a
superior optimum as compared to the basic GA method. The new hybrid PSOGA method is expected
to perform as an effective tool for rapid engine performance optimization.
2015 Elsevier Ltd. All rights reserved.

1. Introduction
Diesel engines have been the primary power source for industrial and transportation applications because of their advantages
in high efficiency and high power density compared to sparkignition (SI) engines [1]. Nonetheless, there are still incentives to
improve the performance of diesel engines, namely reducing the
brake specific fuel consumption (BSFC) and high exhaust emissions
of particulate matter (PM) and oxides of nitrogen (NOx) [25]. On
the other hand, in order to reduce the dependence on fossil fuels,
many different types of alternative fuels have been under investigation [68]. Biodiesel, a mixture of fatty acid methyl ester that can
be produced from different biomass feedstocks, is a popular alternative to diesel fuel. It has been widely accepted that biodiesel can
effectively reduce engine-out emissions of PM, carbon monoxide

Corresponding author.
E-mail address: kong@iastate.edu (S.-C. Kong).
http://dx.doi.org/10.1016/j.apenergy.2015.12.044
0306-2619/ 2015 Elsevier Ltd. All rights reserved.

(CO), and unburned hydrocarbons (HC), but combined with an


increase in NOx emissions [913]. Another important advantage
of biodiesel is that it can be used in an engine without the need
for major modification to the engine set-up. Biodiesel is often
blended with regular diesel fuel for use in engines. Different factors
can influence the combustion characteristics of diesel engines
fueled with biodiesel [14,15].
In order to explore a favorable in-cylinder combustion process
when using biofuel to meet the strict emission regulation, its
important to create a controllable mixture prior to ignition by
the proper organization of the engine operating parameters
[1620]. Some statistical tools such as artificial neural network
(ANN), response surface methodology (RSM) and Taguchi method
[2124] are used to obtain the optimal operating conditions. However, the accuracy of the results using these methods is not always
guaranteed when dealing with small sample database and the optimization process is timing consuming. In recent years, Genetic
Algorithm (GA) or Particle Swarm Optimization (PSO) [25], the

Q. Zhang et al. / Applied Energy 165 (2016) 676684

typical random methods by using probabilistic calculations to find


the optimal values, have become increasingly popular for engine
optimization [2632]. These methods have been confirmed when
they are formulated properly. GA is numerous in construction
and models the Darwinian theory of evolution. The best solutions
in a given set, based on fitness, survive or are selected to breed
and produce new solutions incorporating components of their parents based on crossover of genes. As in the evolution of species random mutations can be applied to subsequent generations to
enhance the search. PSO is based on the social and cooperative
behavior shown by various species like flocks of birds or schools
of fish. PSO is initialized with a population of potential solutions
(called particles). The particles move through the search space
with a specified velocity. Each particle maintains a memory about
the track of its previous best position. By this way, the particle
adjusts its position for the optimal solution in the search space
according to its own flying experience and that of its neighbors.
It is well known that the engine control parameters such as EGR
rate, fuel pressure, main SOI, pilot timing and pilot rate have
greatly effect on engine performances (exactly the emission of
PM and NOx) [24,32]. These parameters are in conflict with each
other and finding the best trade-offs between strategies to optimize the engine performance can be time consuming. EGR recycles
exhaust gas to the intake manifold, increasing the specific heat of
the charge gas. This has the effect of reducing maximum incylinder temperatures leading to a reduction in NOx. EGR phases
combustion to later crank angles and can inhibit mixing between
air and fuel resulting in increased emissions of incomplete combustion [24,32]. Retarding injection timing to later in the cycle
can also reduce peak temperatures while increasing BSFC [24].
Pilot injections and fuel pressure can be used to enhance mixing
of air and fuel at the cost of increased NOx emissions without
EGR [32]. Therefore a careful balance between mixing and temperature reduction strategies is required to reduce overall emissions
simultaneously.
For these reasons, a modified strategy for PSO and GA is
proposed which combines the exploitation of PSO with the exploration of GA. For engine performance optimization, the mechanics
based on a small population (e.g. micro GA [29]) are of great
interest, indicating the little requirement in resources for engine
testing. In PSO, particles in the swarm share information between
them while at the same time they have the tendency to move randomly, which speeds up the search processing. However, when
dealing with the problems with small population, PSO may not
retain enough diversity to converge to the global optimum. The
key genetic operators in GA, namely selection, crossover and mutation may compensate this shortage. Hybrid algorithms which
attempt to combine the advantages of different algorithms are
investigated to improve optimization efficiency [3339]. The key
distinction of the proposed hybridization of PSO and GA is to
exploit the complementary characteristics of both optimization
strategies decreasing search time to the global optimum, which
will minimize the cost of engine experiments. The search time
here means the number of test function calculations to find the
best solution, i.e., the number of engine runs to find the favorable
operating condition.
The objective of the present study is to develop and apply a
methodology that can greatly reduce the number of engine runs
to optimize the operating conditions of a diesel engine. This study
aims to add PSO to the GA process to improve the efficiency of GA
learning. Specifically, the mechanism of hybrid optimization is
implemented to obtain improved control parameter settings of a
diesel engine fueled with biodiesel. This study contributes to the
literature by proposing an optimization method that works in relatively small-scale tests and is expected to be broadly applicable to
engines or other combustion systems.

677

2. Model hybrid PSOGA for optimization


2.1. Algorithm description
This paper presents a hybrid PSOGA algorithm based on a
small population. In the general PSO model, the position and velocity of a particle at the current iteration, X(i) and V(i), respectively,
can be updated according to the following formulas.

Xi 1 Xi Vi 1

Vi 1 wVi C 1 R1 Pbest  Xi C 2 R2 Gbest  Xi

In the above equations, i is the current iteration; i + 1 is the next


iteration; w is the inertia weight and is used to reduce the current
velocity or the impact of the previous velocity. Two weight parameters, C1 (individual interest factor) and C2 (social interest factor)
are used to balance the individual and social confidence factors
for the particle. R1 and R2 are two random numbers between 0
and 1 to enrich the search space. Pbest is the particles local best,
and Gbest is the global best, i.e., the best among all the particles. In
application, X is a design variable, and its value is modified constantly by V. The modification considers the information of both
the particle itself and its neighbors. Finally, an objective function
must be given to evaluate the quality of a position. This behavior
continues until a satisfaction criterion is reached.
The continuous representation method of the variables with
single-point crossover is applied because of the high precision, less
storage requirement, and fast execution. Inspired by previous studies [4042], the new offspring generated by GA and their parents
are combined and sorted. Thereafter, the best N members are chosen to form the population of the next generation. It provides the
chance of survival for parents in the next generation, hence the
overall rapid convergence are enhanced. In addition, in the later
process of the evolution, a large variance of variables will result
in the long convergence history. Thus, the dynamic perturbations
of variables needed to be mutated in the population are assigned
according to Eq. (3) rather than a fixed value within the variables
bounds.

X m i 1 X c i 1 rand=i

In Eq. (3), rand is a random number in the range of 1 to 1, and i is


the number of iteration. Xc(i + 1) is the value after the crossover
operation in GA.
The present hybrid method is done in a two-step process. A
generation has N particles and their values are evaluated. First, a
PSO update step will be done on the best n particles and produce
n offspring for use in the next generation. n is determined by multiplying the keep rate (that is n/N) by the total number of variables.
The rest Nn particles will be discarded to make room for the new
offspring generated by the GA step. Second, in the GA step,
genetic variants are taken by crossing over with the survived n
PSO particles and dynamic mutation to generate Nn members.
As a result, the new generation will have N particles whose values
will be evaluated again. The flowchart showing the hybrid PSOGA
is presented in Fig. 1.
Overall, the hybrid method constructs a PSOGA stepwise algorithm based on a PSO update step followed by a GA step. The
procedure can be detailed as follows.
Initialization: Set up the required parameters.
PSO update steps:
Step 1: Set the number of iteration i. Initially i is equal to 0. Randomly generate the initial positions X(i) = {xj(i)} and
velocities V(i) = {vj(i)} of particles, where j is the index
of the particles, j = (1, 2, . . ., N) and N is the size of
swarm.

678

Q. Zhang et al. / Applied Energy 165 (2016) 676684

Generation i

P1

Pn

Select the best n


members

Pn+1

PN

Discard the worse


N-n members

Update position and velocity


and create n new members
PSO Steps
P1

...

Pn

Tournament selection

Crossover

Dynamic mutation

GA Steps

Create N-n new


members
Pn+1

...

PN

Combine PSO and


GA members

Generation i+1

P1

Pn

Pn+1

PN

Fig. 1. Flowchart of the hybrid PSOGA algorithm.

Step 2: Calculate the value of the objective function. If the termination condition is met, the algorithm terminates.
Step 3: Obtain the new velocities V(i + 1) and positions X(i + 1)
of particles using Eqs. (1) and (2), and then update
Pbest,j and Gbest.
Step 4: Identify the best n members and discard the rest of the
Nn members.
GA steps:
Step 5: Tournament selection based on n members. Select these
n individuals from X(i + 1) to form the mating pool with
a population of Xs(i + 1).
Step 6: Crossover. Perform crossover operation on population
Xs(i + 1) to form a population Xc(i + 1).
Step 7: Dynamic mutation. Mutate a single element of an individual with the mutation rate of pm to form a population
Xm(i + 1) using Eq. (3) and create Nn members.
Step 8: Form the new generation i + 1 which includes n
members from PSO and Nn members from GA.
Return to step 2.
2.2. Testing hybrid PSOGA for benchmark functions
Three widely used multimodal minimization benchmark functions, listed in Table 1, were used to test the performance of the

present hybrid PSOGA algorithm. Finding the global minimum


of these functions is a fairly difficult problem due to their large
number of local minima. Since the optimization of engine performance always deals with multiple parameters, the number of variables for the simulation was set to five. The keep rate was set at 0.5
by a trial-and-error method, meaning that half of the population
survive and become potential parents. In the GA steps, the crossover probability is 1.0 and dynamic mutation is applied. The
parameters C1, C2 and w in PSO are set to 2.0, 2.0 and 0.7298,
respectively, as suggested in the literature [32,33].
In order to make sure that the hybrid PSOGA is efficient with
the use of a small population, the population size was set to 6, 8,
and 10 in 3 tests. The convergence histories of the 3 different test
functions for 100 times with different initial populations were
obtained. The maximum generation is set as 100 for each time.
The average error between the known global optimum and the
best successful points found were calculated and the averages of
function evaluations are shown in Table 2.
Table 2 shows that a population of 10 and 8 resulted in the fastest convergence for the Ackleys function during the search process, in which the average number of function evaluation was
the same (40). The average error with a population of 8 and 10 is
also smaller than that of population of 6. The average numbers of
function evaluations to achieve the minimum of the Rastrigins
function were 54, 42 and 43 for the population size of 6, 8 and
10, respectively. It was found that the average error was 2.0612

679

Q. Zhang et al. / Applied Energy 165 (2016) 676684


Table 1
Details of benchmark functions [43].
Function

Equations
q
Pn

Ackley

x2
1 i

Pn

cos2pxi

n
n
f 1 x 20exp0:2
 exp

P 
f 2 x n1 x2i  10 cos2pxi 10
 
P  x2i  Qn
 1 10 cos pxii 1
f 3 x n1 4000

Rastrigin
Griewangk

Range of variables

Optimum

[32.768, 32.768]

[5.12, 5.12]

[600.0, 600.0]

20 e

Table 2
Results of hybrid PSOGA with population 6, 8 and 10 for three test functions.
Test function

Average of function evaluation numbers

f1
f2
f3

Average error

N=6

N=8

N = 10

N=6

N=8

N = 10

50
54
58

40
42
46

40
43
44

1.2072e08
2.0612
0.19958

4.0966e14
0.08266
0.028726

1.0694e14
0.056515
0.0219163

when using the population size of 6. In the case of Griewangks


function, it required 58, 46 and 44 times of function evaluations
for the hybrid method to find the global minimum using a population of 6, 8 and 10, respectively. In general, the algorithm with a
population of 6 converged slowly. There was no distinctive difference between the population size of 8 and 10 in most cases. As a
result, an appropriate size of population was chosen as 8 as small
population is desired to minimize the overall testing duration.
Numerical simulations based on the basic GA (BGA) and basic
PSO (BPSO) were also conducted using a population of 8, the
results are listed in Table 3. It can be seen that the average errors
for hybrid PSOGA are smaller than those of BGA and BPSO from
Table 3. In the case of BPSO, there are no GA operators in the optimal processing, the average errors for three functions are greater
than PBSO and PSOGA, the algorithm seems to easily fall into
the local optima. In the case of BGA, the average error is in the middle among the three methods. In all, the performance of the PSO
GA is more efficient. The hybrid PSOGA algorithm proposed is
more competitive than BGA and BPSO in the view of the algorithms ability to find improved results in less time. In this work,
the focus is centered on the comparison of hybrid PSOGA and
BGA, the detailed comparison of the hybrid PSOGA and BPSO
can be found in previous studies [39].
3. Experimental setup
3.1. Engine facilities
The experiments were conducted on a four-cylinder, fourstroke turbocharged John Deere 4045T diesel engine with a highpressure common-rail injection system. The engine specifications
are given in Table 4, and soy biodiesel was used as the test fuel.
The fuel injection parameters were controlled using the John Deere
DevX for injection pressure, injection timing, number of injections,
and fuel quantities. The EGR system comprised of an EGR cooler
and a pump that was driven and controlled externally by a
three-phase motor. The pump speed was used to control the mass

Table 3
Results of PSOGA, BGA and BPSO for three functions.
Test function

f1
f2
f3

Table 4
Test engine specifications.
Type of engine

John Deere 4045T

Cylinder number
Bore (mm)
Stroke (mm)
Compression ratio
Injection system
Intake valves/exhaust valves

4
106
127
17.0:1
Common rail
2/2

flow rate of EGR from the exhaust. The cylinder pressure was measured using a Kistler 6125A pressure transducer and a Kistler
5010A charge amplifier. The data was processed using a customized LabVIEW program which captured and averaged cycle
data for cylinder pressure and heat release rate analysis.
The exhaust emissions were measured using a Horiba MEXA
7100 DEGR gas analyzer for NOx, HC, CO2, CO and O2. The primary
sampling and secondary sampling allowed for a live EGR calculation using the CO2 ratio between the intake and exhaust gases.
The particulate matter emissions were measured using an AVL
415S smoke meter.
The load condition chosen was approximately 50% of the rated
constant-duty load at 1400 rpm, which represents an important
part-load condition where control parameters play a critical role
in exhaust emissions and fuel economy. On each testing day, the
engine was started from rest and allowed to run for at least
30 min for warm up. The conditions of each test were determined
by the optimization algorithm. Following the input of each test
points operation parameters the engine was allowed to run for
at least ten minutes to assure steady state operation. In total, five
variables were changed throughout the experiments that have
non-separable effects on engine performance and emissions. The
test conditions and limits of the engine control parameters are
shown in Tables 5 and 6.

Table 5
Test conditions.

Average error
PSOGA

BGA

BPSO

4.0966e14
0.08266
0.028726

1.2415
0.60969
0.30626

1.3279
1.1741
1.5254

Speed (rpm)
Torque (N-m)
Intake temperature (C)
EGR temperature (C)
Fuel temperature (C)

1400
149
40
40
16.3

680

Q. Zhang et al. / Applied Energy 165 (2016) 676684

Table 6
Engine control parameter limits.
Parameter

Minimum

Maximum

EGR (%)
Fuel pressure (MPa)
Pilot timing (CAD ATDC)
Pilot ratio (%)
Main SOI (CAD ATDC)

2.0
113.0
40.0
2.0
15

50.0
200.0
0
65.5
5

3.2. Objective function and its evaluation


The objective function for the minimization problem considered
NOx, PM, HC, CO, and brake specific fuel consumption (BSFC). The
utopia point, based on the 2014 U.S. EPA Tier 4 off-road standards,
was chosen to reflect the design goal. The overall objective function value, denoted as Fobj, is shown in Eq. (4), with ideal values
taken from Table 7. The Fobj decreases with the reduction in emissions and BSFC. The first item in right hand of Eq. (4) is designed in
terms of extraction of square root to well weight the great trade-off
between NOx and PM, which can vary a lot in Fobj when there is the
little change in NOx or PM value. In order to monitor the evolutions
of NOx and PM during optimization, objective functions that only
consider one pollutant were also formulated in non-dimensional
form, as shown in Eqs. (5) and (6). These two objective functions
will be used for assessing the direct impacts of operating conditions on PM and NOx emissions alone.

2 
2 #0:5
NOxmeas
PMmeas

NOxideal
PMideal

 
 


COmeas
HC meas
FC meas

3  COideal
3  HC ideal
3  FC ideal


NOxmeas
NOxobj
NOxideal


PMmeas
PMobj
PM ideal

Fig. 2. Values of various objective functions using the basic GA method.

"

F obj

4
5

Fig. 3. PM and NOx objective values using the basic GA method.

4. Results and discussion


4.1. Basic GA
Optimization using the basic GA method was performed for ten
generations, with each generation consisting of eight particles (i.e.,
operating conditions). Fig. 2 shows the evolutions of different
objective functions during the optimization. The results using the
basic GA method showed an overall decrease in the objective function value. The tendency of the different objective values was
toward the clumping until the minimum appeared to remain near
stagnant after the fifth generation.
Fig. 3 shows the evolution of the best and average PM and NOx
objective values. PM and NOx emissions were improved at the later
generations but the trade-off showed little deviation. The optimal
NOx-PM solutions remained nearly unchanged from generation 5
to generation 8. Overall, the optimal objective values for PM and
NOx are 1.18 and 4.25, respectively.

Table 7
Ideal values for emissions.
Emission objective

Tier 4 regulation

Ideal values

NOx (g/kW h)
NMHC (g/kW h)
CO (g/kW h)
PM (g/kW h)
BSFC (g/kW h)

0.40
0.19
5.0
0.02

0.20
0.19
5.0
0.01
200

Fig. 4. Values of the overall objective function using the basic GA method.

The overall improvement by use of GA can be seen in Fig. 4. The


objective values varied significantly within the same generation in
the early generations until the sixth generation. It indicates the
ability of exploration inherited by the GA method. On the other
hand, this also means a high chance of the solution staying away
from the optimal area, e.g., the first particle in the fifth generation
in Fig. 4. This feature can result in slow convergence.
Fig. 5 presents the progression of the minimum objective value
found and the average value of the particles by generation. It illustrates the discontinuous tendency in average values between generations while the relatively continuous improvement in the best
value of population. The minimum value found after 10 generations was 5.45 and the final average of the particles was 5.71.

Q. Zhang et al. / Applied Energy 165 (2016) 676684

Fig. 5. Average and minimum values of the overall objective function using the
basic GA method.

681

Fig. 7. Average and minimum values of the objective function for NOx and PM
using hybrid PSOGA method.

4.2. PSOGA hybrid


In this study, six generations were evaluated and each
generation used eight particles. Fig. 6 shows the values of different
objective functions during optimization. Results using the hybrid
PSOGA method showed a rapid decrease in the average objective
function value, indicating a tendency toward rapid clumping. This
is due to the PSO methods aggressive acceleration toward the
optimum as a whole while early exploration was inhibited by
the strong drive toward one optimum value. Clearly evident was
the ability of the hybrid method to maintain exploration and
diversity into the later evaluations after Generation 3 due to the
combined overshoot of the PSO and the diversity influence of the GA.
Fig. 7 illustrates that the hybrid method was able to produce
substantially low levels of PM and NOx at each generation, even
as convergence was approached. This illustrates the ability of the
GA step in the hybrid method to maintain diversity and aggressive
searching even as the drive of the PSO was reduced near the optimum. As shown in Fig. 7, both the average and minimal objective
values for PM and NOx were reduced continuously. The optimum
objective values found for PM and NOx were 0.74 and 3.71,
respectively.
The overall improvement achieved by the hybrid method can be
seen in Fig. 8. The hybrid method consistently produced the low
overall objective values from generation to generation. In the early
generations, the downtrend of the overall objective values was
clear until the particles approached the finial minimum.
Fig. 9 shows the progression of the best value found and the
average by generation, which shows that both the minimum and
the average decreased consistently during the optimization. In

Fig. 8. Values of the overall objective function using the hybrid PSOGA method.

Fig. 9. Average and minimum values of the overall objective function using hybrid
PSOGA method.

the first two generations, the spread of the objective values with
respect to the average was approximately one times the minimum
value. After that, the fluctuation of the difference between the minimum and the average was small and reached to zero, indicating
the final optimal solution was found. The best value found after
six generations was 4.71 with the average value of 4.79.
4.3. Comparison of basic GA and PSOGA

Fig. 6. Values of various objective functions using the hybrid PSOGA method.

The hybrid method was compared with the basic GA method,


and the relative reductions in PM, NOx, and overall objective
values were examined. The hybrid method has the advantages of
fast convergence and good balance of exploration and exploitation
when compared with the basic GA, as shown in Fig. 10. The

682

Q. Zhang et al. / Applied Energy 165 (2016) 676684

variation between the minimum and average objective values in


each generation for the hybrid method was smaller than those
for the basic GA method, especially in the middle of the optimization,
indicating that the hybrid method can speed up the convergence.
It can be seen that the best overall objective value was better for
the hybrid method than the basic GA, which is attributed to the
ability for the hybrid method to find the global minimum.
Figs. 11a and 11b show a continuous reduction in the PM and
NOx objective values, especially a significant reduction in the lowest PM objective value when the hybrid method was employed. A
38% reduction in the best PM objective value and a 13% reduction
in the best NOx objective value were obtained by use of the hybrid
method as compared to the basic GA. It indicates that the hybrid
method can explore the solution space more efficiently in which
some operating parameters were highly sensitive to the pilot injection parameters but easily discharged by the GA method. In addition to the obvious reduction of the two most critical emissions
from diesel engines, the final best value found by PSOGA has a
reduction of 25% in HC emissions, 5.5% in CO emissions, and 2.6%
in the BSFC, as compared to the basic GA, as also shown in Tables
8 and 9.

Fig. 11a. Minimum of the PM objective function using basic GA and hybrid
PSOGA.

4.4. Comparison of engine test results


The actual dynamometer time is a vital factor to evaluate the
effectiveness and efficiency of an algorithm for engine performance
optimization. In the present study, a minimum of ten minutes was
given for each test point to ensure steady-state conditions. When
the basic GA method was used, due to the significant variation
between the operating points in the same generation, some tests
required more than ten minutes to reach the steady state. The
hybrid method and basic GA method required 48 and 80 engine
tests, respectively, to reach their respective optima. It is clear that
the hybrid method will find the best solution with fewer points
and with nearly half of the dynamometer time. The final optimal
parameters and emissions results using the basic GA and the
hybrid method are listed in Tables 8 and 9.
Fig. 12 shows the comparison of the in-cylinder pressures for
the best solutions of generation 2, 3, 4 and 6 using the hybrid
PSOGA, along with the corresponding heat release rates. It is of
interesting to note that the difference between the pilot timing
and main SOI remained the same at 4.91 CAD for each optimum
after generation 2. The other four parameters changed through
the evolution. The pilot timing moved closer to TDC and the pilot
percentage decreased from 47 to 44.9 percent while EGR was
increased. With the same offset in the injection timing for each
optimum, the main SOI moved beyond TDC, thus lowering the
maximum cylinder pressure. The best condition in generation 2
(point 16) still exhibited the typical diesel combustion characteristics

Fig. 10. Minimum and average objective values using basic GA and hybrid PSOGA.

Fig. 11b. Minimum of the NOx objective function using basic GA and hybrid
PSOGA.

Table 8
Best operating conditions for selected generations of hybrid PSOGA.
Parameter

Gen #2
Point 16

Gen #3
Point 24

Gen #4
Point 32

Gen #6
Point 48

EGR (%)
Fuel pressure (MPa)
Pilot timing (ATDC)
Pilot ratio (%)
Main SOI (ATDC)

43.5
175.6
8.19
47.0
3.28

48.8
174.2
3.19
45.4
1.72

49.1
173.9
2.17
45.0
2.74

49.9
173.9
1.81
44.9
3.10

Table 9
Optimal operating parameters and corresponding emissions found.
Parameter

Optimal value
by basic GA

Optimal value by
hybrid PSOGA

EGR (%)
Fuel pressure (MPa)
Pilot timing (ATDC)
Pilot ratio (%)
Main SOI (ATDC)
NOx (g/kW h)
HC (g/kW h)
CO (g/kW h)
PM (g/kW h)
BSFC (g/kW h)

48.7
172.1
2.06
41.7
3.11
0.850
0.190
3.08
0.0120
296.8

49.9
173.9
1.81
44.9
3.10
0.742
0.142
2.91
0.00742
289.1

(i.e., distinct premixed and diffusion combustion). However, the


best conditions beyond generation 3 exhibited highly premixed
combustion with slight cool flame heat release between 0 and 5
ATDC. Furthermore, an increase in the EGR rate from 43.5 to 49.8

Q. Zhang et al. / Applied Energy 165 (2016) 676684

683

because of the PSO methods aggressive acceleration toward the


optimum as a whole and at the same time GAs search for the
diversity of the solution space. The hybrid PSOGA method is capable of finding highly sensitive operating parameters which could
be easily ignored in a parametric study on optimizing diesel engine
performance. Based on the results of this study, the new hybrid
PSOGA methodology is expected to be an effective tool for rapid
optimization of engine control parameters.
Acknowledgment

Fig. 12. Comparison of in-cylinder pressure and heat release rate between each
generation for the hybrid PSOGA.

The authors acknowledge the support by John Deere Power Systems for this research.
References

Fig. 13. Comparison of the optimal in-cylinder pressure and heat release rate for GA
and hybrid PSOGA.

percent has decreased NOx emissions. Generation 6 has a slightly


lower maximum cylinder pressure due to the late injection and
high EGR.
Fig. 13 compares the cylinder pressures and heat release rates
for the optima found by the hybrid PSOGA and basic GA. The
results are very similar. Table 9 shows that the operating conditions were also very similar. Overall, the optimal operating conditions required a large pilot injection near TDC and high EGR rates.
It is clear in Fig. 13 that there was a short ignition delay of the
pilot fuel between the injection time (approximately 2 ATDC)
and the heat release (1 ATDC). The injection of the pilot fuel is
shown as a slight increase in the heat release rate at 1 ATDC. The
main heat release started at 4.5 ATDC as illustrated by a rapid
increase in pressure and heat release rate. The peak heat release
for the PSOGA optimum was slightly higher than the GA optimum
because of the large and late pilot injection of the hybrid PSOGA
optimum. This can be the result of more complete oxidation of
the fuel, leading to lower hydrocarbon and carbon monoxide emissions. It is possible that the increased amount of pilot fuel improved
the in-cylinder airfuel mixing and thus extended the premixed
combustion region, which also aided in the reduction of PM. In general, the hybrid PSOGA method can generate a superior solution
than the basic GA, proving that there were operational points that
were highly sensitive to the pilot injection parameters. These pilot
parameters would have easily been ignored in a common parametric study without properly exploring the search space.
5. Conclusion
The modified hybrid PSOGA approach is clearly an improvement over the basic GA method. Further, the hybrid method shows
improved performance against BGA and BPSO in standard benchmark tests. The results of this study indicate that the modified
hybrid approach offered faster convergence and wider spread

[1] Reitz RD. Directions in internal combustion engine research. Combust Flame
2013;160:18.
[2] Reiter AJ, Kong SC. Demonstration of compression-ignition engine combustion
using ammonia in reducing greenhouse gas emissions. Energy Fuels
2008;22:296371.
[3] Suh HK. Investigations of multiple injection strategies for the improvement of
combustion and exhaust emissions characteristics in a low compression ratio
(CR) engine. Appl Energy 2011;88:50139.
[4] Gonca G, Sahin B, Parlak A, Ust Y, Ayhan V, Cesur I, et al. Theoretical and
experimental investigation of the Miller cycle diesel engine in terms of
performance and emission parameters. Appl Energy 2015;138:1120.
[5] Squaiella LLF, Martins CA, Lacava PT. Strategies for emission control in diesel
engine to meet Euro VI. Fuel 2013;104:18393.
[6] Roy MM, Wang W, Bujold J. Biodiesel production and comparison of emissions
of a DI diesel engine fueled by biodieseldiesel and canola oildiesel blends at
high idling operations. Appl Energy 2013;106:198208.
[7] Sukuraman S, Kong SC. Numerical study on mixture formation characteristics
in a direct-injection hydrogen engine. Int J Hydrogen Energy
2010;35:79918007.
[8] Zhang L, Kong SC. Multicomponent vaporization modeling of bio-oil and its
mixtures with other fuels. Fuel 2012;95:47180.
[9] zener O, Yksek L, Ergen AT, zkan M. Effects of soybean biodiesel on a DI
diesel engine performance, emission and combustion characteristics. Fuel
2014;115:87583.
[10] Labecki L, Cairns A, Xia J, Megaritis A, Zhao H, Ganippa LC. Combustion and
emission of rapeseed oil blends in diesel engine. Appl Energy 2012;95:13946.
[11] Tesfa B, Mishra R, Zhang C, Gu F, Ball AD. Combustion and performance
characteristics of CI (compression ignition) engine running with biodiesel.
Energy 2013;51:10115.
[12] Kuzhiyil N, Kong SC. Energy recovery from waste plastics by using blends of
biodiesel and polystyrene in diesel engines. Energy Fuels 2009;23:324652.
[13] Varuvel EG, Mrad N, Tazerout M, Aloui F. Experimental analysis of biofuel as an
alternative fuel for diesel engines. Appl Energy 2012;94:22431.
[14] Tompkins BT, Song H, Bittle JA, Jacobs TJ. Efficiency considerations for the use
of blended biofuel in diesel engines. Appl Energy 2012;98:20918.
[15] Lenik L, Vajda B, Zuni Z, kerget L, Kegl B. The influence of biodiesel fuel on
injection characteristics, diesel engine performance, and emission formation.
Appl Energy 2013;111:55870.
[16] Reitz RD, Duraisamy G. Review of high efficiency and clean reactivity
controlled compression ignition (RCCI) combustion in internal combustion
engines. Prog Energy Combust Sci 2015;46:1271.
[17] Molina S, Garca A, Pastor JM, Belarte E, Balloul I. Operating range extension of
RCCI combustion concept from low to full load in a heavy-duty engine. Appl
Energy 2015;143:21127.
[18] Karra PK, Veltman MK, Kong SC. Characteristics of engine emissions using
biodiesel blends in low temperature combustion regimes. Energy Fuels
2008;22:376370.
[19] Asad U, Kumar R, Zheng M, Tjong J. Ethanol-fueled low temperature
combustion: a pathway to clean and efficient diesel engine cycles. Appl
Energy 2015;157:83850.
[20] Lu XC, Han D, Huang Z. Fuel design and management for the control of
advanced compression-ignition. Prog Energy Combust Sci 2011;37:74183.
[21] Kalogirou SA. Artificial intelligence for the modeling and control of
combustion processes: a review. Prog Energy Combust Sci 2003;29:51566.
[22] Lee DH, Park JS, Ryu MR, Park JH. Development of a highly efficient lowemission diesel engine-powered co-generation system and its optimization
using Taguchi method. Appl Therm Eng 2013;50:4915.
[23] Hirkude JB, Padalkar AS. Performance optimization of CI engine fuelled with
waste fried oil methyl ester-diesel blend using response surface methodology.
Fuel 2014;119:26673.
[24] Molina S, Guardiola C, Martn J, Garca-Sarmiento D. Development of a controloriented model to optimise fuel consumption and NOX emissions in a DI
Diesel engine. Appl Energy 2014;119:40516.
[25] Haupt RL, Haupt SE. Practical genetic algorithms. 2nd ed. New Jersey: WesleyInterscience; 2004.

684

Q. Zhang et al. / Applied Energy 165 (2016) 676684

[26] Donateo T, Tornese F, Laforgia D. Computer-aided conversion of an engine


from diesel to methane. Appl Energy 2013;108:823.
[27] Li YG, Pilidis P. GA-based design-point performance adaptation and its
comparison with ICM-based approach. Appl Energy 2010;87:3408.
[28] Deb M, Banerjee R, Majumder A, Sastry GRK. Multi objective optimization of
performance parameters of a single cylinder diesel engine with hydrogen as a
dual fuel using pareto-based genetic algorithm. Int J Hydrogen Energy
2014;39:806377.
[29] Wickman DD, Senecal PK, Reitz RD. Diesel engine combustion chamber
geometry optimization using genetic algorithms and multi-dimensional spray
and combustion modeling. SAE technical paper 2001-01-0547; 2001.
[30] Shi Y, Reitz RD. Optimization of a heavy-duty compression-ignition engine
fueled with diesel and gasoline-like fuels. Fuel 2010;11:341630.
[31] Mehrani P, Watson HC. Joint efficiency and NOx optimization using a PSO
algorithm. SAE technical paper 2006-01-1109; 2006.
[32] Karra PK, Kong SC. Application of particle swarm optimization for diesel
engine performance optimization. Combust Sci Tech 2010;182:879903.
[33] Wong KI, Wong PK, Cheung CS, Vong CM. Modeling and optimization of
biodiesel engine performance using advanced machine learning methods.
Energy 2013;55:51928.
[34] Wei ZB, Li XL, Xu LJ, Cheng YT. Comparative study of computational
intelligence approaches for NOx reduction of coal-fired boiler. Energy
2013;55:68392.

[35] Yang J, He LF, Fu SY. An improved PSO-based charging strategy of electric


vehicles in electrical distribution grid. Appl Energy 2014;128:8292.
[36] Mena R, Hennebel M, Li YF, Zio E. Self-adaptable hierarchical clustering
analysis and differential evolution for optimal integration of renewable
distributed generation. Appl Energy 2014;133:388402.
[37] Ganesan T, Elamvazuthi I, Shaari KZK, Vasant P. Swarm intelligence and
gravitational search algorithm for multi-objective optimization of synthesis
gas production. Appl Energy 2013;103:36874.
[38] Martinez-Rojas M, Sumper A, Gomis-Bellmunt O, Sudri-Andreu A. Reactive
power dispatch in wind farms using particle swarm optimization technique
and feasible solutions search. Appl Energy 2011;88:467886.
[39] Bertram AM, Zhang Q, Kong SC. A novel particle swarm and genetic algorithm
hybrid method for diesel engine performance optimization. Int J Engine Res
2015.
[40] Deb K, Pratap A, Agarwal S, Meyarivan T. A fast and elitist multiobjective
genetic algorithm: NSGA-II. IEEE Trans Evolut Comput 2002;2:18297.
[41] Basu M. Fuel constrained economic emission dispatch using nondominated
sorting genetic algorithm-II. Energy 2014;78:64964.
[42] DErrico G, Cerri T, Pertusi T. Multi-objective optimization of internal
combustion engine by means of 1D fluid-dynamic models. Appl Energy
2011;88:76777.
[43] Yang XS. Nature-inspired optimization algorithms. MA: Elsevier; 2014.

Вам также может понравиться