Applied Energy 187 (2017) 601611

Contents lists available at ScienceDirect

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

Multi-objective optimization of ethanol fuelled HCCI engine

performance using hybrid GRNNPSO
Harisankar Bendu a,, B.B.V.L. Deepak b, S. Murugan a
a
Department of Mechanical Engineering, National Institute of Technology Rourkela, India
b
Department of Industrial Design, National Institute of Technology Rourkela, India

h i g h l i g h t s

Novel GRNNPSO model for the rapid optimization of the HCCI engine.

At 170 C, a maximum of 45% thermal efficiency is found for ethanol HCCI engine.

Negligible NOx and smoke are observed for port injected ethanol HCCI operation.

The developed tool is flexible for any user defined weights for optimization.

The proposed model obtained the optimized HCCI engine parameters within 75 ms.

a r t i c l e i n f o a b s t r a c t

Article history: In this research study, the ethanol-fuelled homogeneous charge compression ignition (HCCI) engine oper-
Received 23 August 2016 ating conditions were optimized based on its performance and emission parameters. However, there is no
Received in revised form 7 November 2016 proper relation among these performance and emission parameters. Therefore, it is a challenging task to
Accepted 18 November 2016
optimize the HCCI engine. For this purpose, a hybrid generalised regression neural network (GRNN)par-
ticle swarm optimisation (PSO) model was designed to optimize three input parameters, including the
charge temperature, engine load, and EGR rate. The GRNN based prediction tool has been introduced
Keywords:
to obtain the relation between input and output parameters. Initially, PSO generated a random particle
HCCI
PSO
and given as input to GRNN. The GRNN carries a nonlinear regression analysis among all the input and
Ethanol output parameters of the ethanol fuelled HCCI engine. The optimum kernel bandwidth in the GRNN
GRNN model was determined by the grid search method for reducing the cross-validation error. The output
GRNNPSO of the GRNN regarding performance and emission parameters was given to the PSO algorithm to optimize
the HCCI engine operating conditions. The developed algorithm has the flexibility to optimize the engine
performance for any user-defined weights. The optimum HCCI engine operating conditions for the gen-
eral criteria were found to be 170 C charge temperature, 72% engine load, and 4% EGR. The proposed
model obtained the optimized HCCI engine input parameters within the short span of time i.e., 60
75 ms. The new hybrid GRNNPSO algorithm is anticipated to accomplish as a useful tool for a rapid
engine optimization.
2016 Elsevier Ltd. All rights reserved.

1. Introduction
Automobiles are one of the essential parts of our daily life for
both private and public transportation. The increase in the vehicle
population has a great impact on the human life, such as environ-
mental pollution, climate change, energy crisis, and traffic jam.
Thus, the emission legislations become more stringent with time.
Also, the environmental protection agency (EPA) regulates the
Corresponding author.
E-mail address: hsbendu@gmail.com (H. Bendu).
carbon dioxide (CO2) emissions from the vehicles to improve fuel
economy standards. Simultaneous mitigation of the engine
exhaust emissions with an improved brake specific fuel consump-
tion (BSFC) is a challenging task for researchers and automobile
manufacturers. In the context of clean burning and high efficient
engines, homogeneous charge compression ignition (HCCI) com-
bustion technology is expected to meet the current demands to
increase thermal efficiency and to reduce emissions [1]. In HCCI
engines, the combustion of air/fuel mixture is controlled by the
chemical kinetics mainly dependent on timetemperature history.
The strategy of low-temperature combustion (LTC) is adopted in

http://dx.doi.org/10.1016/j.apenergy.2016.11.072
0306-2619/ 2016 Elsevier Ltd. All rights reserved.
602 H. Bendu et al. / Applied Energy 187 (2017) 601611
the HCCI combustion to reduce trade-off emissions of nitrogen oxi-
des (NOx) and soot simultaneously in the engine cylinder [26].
The LTC reduces the heat losses from the cylinder that results in
a higher brake thermal efficiency (BTE). This lower temperature
will increase the carbon monoxide (CO) and unburned hydrocar-
bon (UHC) emissions.
The main challenges of the engine design engineers are to bal-
ance the NOxPMBSFC trade-off behavior and GHG emissions,
to meet the growing emission standards. One of the possible solu-
tions is to use alternative renewable fuels in the HCCI engines.
Among all the available liquid alternative fuels, ethanol is domes-
tically produced, and has particulate-free burning characteristics.
Ethanol can potentially reduce the global carbon dioxide emissions
[7,8]. Recently, the use of ethanol and methanol in an HCCI engine
by varying the intake air temperatures were experimentally inves-
tigated by Maurya and Agarwal [9]. They have reported that the
methanol and ethanol are the best replacements for the petroleum
fuels. Bahri et al. [10] have predicted the misfire in the ethanol-
fuelled HCCI engines with 100% accuracy, using the artificial neural
network (ANN). Viggiano and Magi [11] have employed a multidi-
mensional CFD code to examine the suitability of ethanol fuel in
the HCCI engines. They have reported high UHC and CO emissions
at lower engine loads. The auto-ignition chemistry of an ethanol
fuelled HCCI engine with the addition of exhaust gas recirculation
(EGR) was studied by Sjberg and Dec [12]. They have reported
that the HCCI combustion of ethanol exhibited a single stage igni-
tion. Also, the oxygen concentration is lowered with the EGR addi-
tion. The addition of medium EGR rate to the LTC diesel engine
potentially reduced the NOx and fuel consumption [13]. Saxena
et al. [14] have made an attempt to recover the waste heat in an
HCCI engine exhaust to decrease the input heat energy for preheat-
ing the wet ethanol. Maurya and Agarwal [15] have also prepared a
homogeneous air/fuel mixture with preheated intake air and port
fuel injection (PFI) techniques. They have found that there was a
stable HCCI operation of ethanol achieved for equivalence ratios
of 2.05.0 at 1500 rpm. Vincenzo [16] optimized performance of
the turbocharged spark ignition (SI) engine using variable valve
actuation (VVA) for a better combustion phasing and a reduced
mixture over-fuelling.
The artificial neural networks (ANN) validates the experimental
results using a differential evolutionary approach that potentially
decreases the computational time due to the complete absence
of complex multiple mathematical equations for solving the
multi-dimensional and non-linear systems. While training, the
ANN can forecast the available experimental data that could be val-
idated with the set of independent data. The ANN has the capabil-
ity to improve the performance of the engine for any of the new
data [17,18]. These ANN models are widely used for prediction of
the conventional engines performance and emission characteris-
tics [1925]. Rezaei et al. [26] have developed feed forward (FF)
and radial basis function (RBF) ANN models for the prediction of
the engine performance metrics with butanol and ethanol fuels.
An error of less than 4% error was obtained with the validated
results. Kara and Baysec [27] predicted the specific fuel consump-
tion (SFC) and torque of a petrol engine with ANN by considering
81 experimental data points. The ANN based performance and
emission characteristics prediction tool was developed for variable
compression ratio (VCR) diesel engine fuelled with waste cooking
oil at various injection timings [28]. Boliviana and Pea [29] have
used generalized regression neural network (GRNN) for controlling
the vibration from diesel engine based on the cylinder pressure.
Choy and Chen [30] have predicted the start of combustion (SOC)
in an HCCI engine, regulated by an auto-ignition with a
multi-input single-output (MISO) model. Their model took about
2030 ms which could be sufficiently faster for the implementa-
tion of real-time control strategies.
Particle swarm optimization (PSO) is an intelligent optimization
algorithm that can be used to find the global optimal value
invented by Kennedy and Eberhart [31] in 1995. Zhang et al. [32]
conducted the performance optimization of a Soy biodiesel fuelled
diesel engine using a basic genetic algorithm (GA) and hybrid
PSOGA. They found that optimization of engine combustion is
an effective with the hybrid PSOGA tool. Manbachi et al. [33]
used PSO and fuzzification to conserve energy and engine
optimization. They studied six different operating scenarios for a
33-node test feeder in a complete day. Lotfan et al. [34] applied
ANN and non-dominated sorting genetic algorithm (NSGA-II) to
reduce emissions from dual fuel engines. They optimized com-
pressed natural gas (CNG) fuel to reduce CO and NOx simultane-
ously. Mohammadhassani et al. [35] combined ANN and ant
colony optimization (ACO) to predict and reduce exhaust emis-
sions with 294 data sets. They reduced the emissions of NOx and
soot, and also optimized fuel consumption and intake air temper-
ature. Mousa et al. [36] have proposed hybrid PSOGA algorithm
for local search optimization. Mozaffari and Azad [37] have opti-
mized exhaust gas temperature (EGT), and UHC emissions during
cold start with hybrid PSO and fuzzy logic. This heuristic multi-
objective algorithm can solve crucial automotive problems. Roy
et al. [38] have optimized the CNG operated performance and
emission tradeoff behavior to reach Tier 4 emissions with the
inherent design of PSO. Park and Choi [39] developed dual-loop
EGR split strategies to investigate split effects on engine perfor-
mance and emissions, thereby minimized NOx formation and fuel
consumption. Pedrozo et al. [8] conducted an optimization at low
load conditions of the ethanol dual-fuel heavyduty diesel engine
to reduce NOx and soot emissions. Various engine optimization
algorithms with their objectives and parameters considered from
the cited literature are listed in Table 1.
To the best of the authors knowledge, this research work is the
first study to develop a comprehensive hybrid GRNNPSO model
for optimizing the performance and emission parameters of the
ethanol-fuelled HCCI engine. The novelty of this work lies in our
complete description of the hybrid model that considered three
input parameters: charge temperature, engine load and EGR; and
six output parameters: BTE, EGT, UHC, CO, NO, and smoke. Four
different cases with user defined weights were considered to show
the flexibility of the model. Initially, the GRNN tool was developed
for predicting the HCCI engine output parameters, i.e., approximate
the relationship among the engine input and output parameters. A
total of 45 steady state experimental data sets was used to train
the GRNN prediction tool. The experiments were conducted on a
single cylinder, four-stroke, air-cooled diesel engine modified to
run in HCCI mode. The charge temperature was varied from 130
to 170 C in steps of 10 C, while ethanol was injected using port
fuel injection technique. The PSO generated the random particle
within the search space, and the fitness of the particle was evalu-
ated with the GRNN tool. The best fitness was stored by replacing
the global fitness value until the completion of the maximum
number of iterations or termination criteria. The HCCI engine
experimental results and the optimization results of the hybrid
GRNNPSO are presented in this paper.
2. Construction of GRNN model for HCCI performance
prediction
The GRNN algorithm was used for the prediction or control. The
extremely parallel structure in the GRNN makes the network learn
sparse data in one-pass. The essential step involved in the process
of GRNN prediction was the pre-processing of the dataset [46]. The
pre-processing of the dataset includes smoothing, omitting out-
liers, recognizing the missing data, etc. [47]. The advantages of
 H. Bendu et al. / Applied Energy 187 (2017) 601611 603

Table 1
Comparison of different algorithms used in engine optimization.

Reference Algorithm Engine Fuel Objective function Parameters considered

Zhang et al. [32] Hybrid PSOGA Diesel engine Soy biodiesel blends Combustion optimization NOx, PM, HC, CO, and BSFC
Wong et al. [40] Hybrid LS-SVM and ELM Diesel engine Biodiesel blends Blend ratio optimization BSFC, CO, CO2, NOX and PM
Bertram et al. [41] Hybrid PSOGA Diesel engine Diesel NOx and PM reduction NOx, PM, HC, CO, and BSFC
Lotfan et al. [34] ANN and NSGA-II Dual fuel engine Diesel and CNG Optimize gaseous fuel CO and NOx emissions.
Sadeghierad et al. [42] GA and PSO Micro turbine NA Generator design Efficiency and power losses
Mohammadhassani et al. [35] Hybrid ANN and ACO Diesel engine Diesel Optimize fuel NOx and soot
consumption and
intake air temperature
Jeong et al. [43] Hybrid GAPSO Diesel engine Diesel Emission reduction Fuel spray, NOx, soot and CO
Mozaffari et al. [37] Neuro-fuzzy and PSO Petrol engine Gasoline Cold start HC emissions EGT, speed, and HC
Roy et al. [38] ANN assisted PSO Diesel engine CNG and diesel Engine optimization NOx, PM, HC, and BSFC
You et al. [44] Hybrid PSO-BPNN Diesel engine Diesel Fault detection Vibration signals
Cai et al. [45] Hybrid PSOEA NA NA Engine data classification NA
This work Hybrid GRNNPSO HCCI engine Ethanol Engine optimization BTE, EGT, NO, Smoke, HC, and CO

EA Evolutionary Algorithm; BPNN Back Propagation Neural Network; ELM Extreme Learning Machine; LS-SVM Least-Squares Support Vector Machine; NA not
available/applicable.

the GRNN were quick learning, best nonlinear mopping capability, its index vector of the influencing input factors (independent vari-
flexible network structure, high fault tolerance and well robustness ables) X. (1).
[48]. The GRNN was based on the probabilistic functional network Pn h  i
that performs regression [49,50]. The four layers in the GRNN are Y i exp D2i =2r2
i1
b X
input layer, radial/pattern layer, regression/simulation layer, and Y Pn h  i 1
i1 exp Di =2r
2 2
an output layer. In the radial layer, the input training data was
clustered. Thus, the number of neurons in the radial layer consist
of an exactly equal number of input training data sets. The regres- where (Xi, Yi) is a sample of (X, Y), n represents the number of train-
sion layer contains an extra neuron in comparison to the output ing samples, and r is the smoothing parameter. Di is the Euclidean
layer. The function of this extra neuron is to calculate the probabil- distance between X and Xi, which can be calculated by Eq. (2).
ity density function, whereas remaining neurons calculate the out-
D2i X  X i T X  X i 2
puts. The GRNN architecture chooses one approximate function
based on the training data that directly relates the input and the Based on GRNN topological structure, the modeling process of
output parameters. Thus, the GRNN network consumes less train- GRNN model is given in the following steps;
ing time compared to the other iterative training networks [51].
The architecture of GRNN model used in this study is described Step 1: Initialize the index vector (input layer) of the influenc-
in Fig. 1. ing factors of HCCI engine by the input neurons.
The mathematical basis of GRNN model for the prediction of an Step 2: Calculate the output value of each radial/pattern
HCCI engine performance is a nonlinear regression analysis neuron.
between HCCI output parameters and its influencing input param-
eters. Eq. (1) can calculate the regression analysis between the The input index vector can be transferred directly to each radial
HCCI performance parameters (dependent variables) Y relative to neuron by setting the weights to 1. The output of each radial neu-

Input Radial Regression Output

layer layer layer layer
f1(X)
BTE
1 Y1
Charge
temperature f2(X) EGT
X1 Y2
2

NO
Engine f3(X) Y3
load
X2 3
Smoke
. . Y4
. .
EGR . . CO
X3 . . Y5
.
F7(X)
UHC
n Y6

Fig. 1. GRNN architecture used in this study.

604 H. Bendu et al. / Applied Energy 187 (2017) 601611

ron uses radial basis function to calculate its output value by All particles within the swarm are communally shared their
Eq. (3). experience, and there exists only one global best position, ever vis-
ited by any of the individuals as represented in Fig. 2. The choice of
f X; X i expD2i =2r2 3 the best particle/position will be determined according to the
Step 3: Calculate the output value of regression/summation obtained fitness values of every individual in the swarm.
neurons. The particle in swarm with optimal fitness value is called as the
The number of neurons in the regression layer consists of arith- global best position and denoted with X gbest . The selection of X gbest
metic summation (Ss) neurons (equal to output layer neurons) and specifies the end of one PSO-iteration. This procedure will be
weighted summation (Sw) neuron. The arithmetic sum of the radial repeated until it reaches the maximum number of iterations or ter-
outputs and the interconnected summation neuron weight equals mination criteria. The new position and velocity of each individual
to 1. The weighted summation Sw computes the weighted sum of particle was updated according to Eqs. (7) and (8).
the radial layer outputs. The output vector can be calculated by v i z 1 v i z C 1  rand1  X pbest  xi C 2  rand2
Eqs. (4) and (5).
 X gbest  xi 7
X
n
Ss expD2i =2 r
2
4
i1
And xi z 1 xi v i z 1 8
where z = number of iteration; rand1 & rand2 = random parameters
X
n
ranging between 0 and 1; and C1 & C2 are cognitive and social
Sw Y i expD2i =2r2  5
i1
parameters.

Step 4: Determine the prediction values of the HCCI engine per- 3.2. PSO implementation
formance metrics.
The output value of Sw will be divided by the output value of Ss. In this research study, three input parameters of the HCCI
The predicted values can be gained by Eq. (6). engine namely charge temperature, engine load, and EGR were
Pn h  i considered. Therefore, the dimension of each swarm particle was
Y i exp D2i =2r2
b X Sw i1 three that corresponded to each input parameter. Later, each ran-
Y Pn h i 6
i1 expDi =2r
Ss 2 2 dom particle was given to the developed GRNN tool as an input
to obtain six output values as discussed. For the obtained output
The smoothening parameter r needs to be set manually values, the fitness of the particle was to be measured as mentioned
throughout the modeling process. The GRNN prediction accuracy in Section 5. The detailed procedure of developed strategy is shown
and the generalization ability are very sensitive to the setting of in Fig. 3.
the smoothening parameter. The optimal smoothening parameter The developed methodology works as per the following steps:
with a minimum cross-validation error was determined using grid
search method. The determination of optimal r value is discussed Step 1. Initialise a particle position and velocity (position = veloc-
in Section 6. ity) with two dimensions within their input range as given
below:
3. Model hybrid GRNNPSO for optimization
130 C < X1 > 170 C (first dimension as charge
temperature)
The performance optimization of the HCCI engine can be
20% < X2 > 100% (second dimension as engine load)
improved by reducing the GRNN prediction error. The PSO
0 < X3 > 7% (third dimension as EGR rate)
algorithm generates a random particle within the search space,
and the GRNN predicts the outputs. The hybrid GRNNPSO model From the experiment, the minimum temperature required to
can be applied to find the optimal fitness. auto-ignite ethanol was 130 C whereas engine knocking at full
load limited the higher intake temperature, i.e., 170 C. The lower
load of the ethanol HCCI was limited by the emissions of UHC
3.1. Particle swarm optimization (PSO) algorithm
and CO in the tailpipe. A small amount of EGR was used to control
the combustion phasing. The further increase in the EGR levels (7%)
PSO is one of the swarm intelligence technique motivated from
led to significant increase in the UHC and CO emissions.
social nature of birds flocking [31]. The birds in PSO are called as a
swarm and each bird in the swarm is considered as a particle.
Step 2. Send the generated input values to the GRNN tool for pre-
Hence, the population in PSO can be identified as a set
dicting six output values as discussed in Section 2.
S fp1 ; p2 ; p3 ; . . . ; pr g.
Where p1 ; p2 ; p3 ; . . . ; pr represents r number of individual
population existing in the swarm.
These individuals are expected to change their position within
the workspace. During the movement of the particles, their new
positions are renewed with an appropriate velocity. Consider r
particles with the positions of fx1 ; x2 ; x3 ; . . . ; xr g with their
corresponding velocities fv 1 ; v 2 ; v 3 ; . . . ; v r g in the swarm. The
new position and velocity of every individual can be achieved
through the conversed information of individuals within the
swarm. This can be achieved in terms of commemoration i.e. each
individual pillories its best position that, it has ever visited during
its search. The best position obtained by every individual is called
position best and is denoted by X pbest . Therefore, there exists r
number of X pbest values for r individuals in the swarm. Fig. 2. PSO structure for its global best approximation.
 H. Bendu et al. / Applied Energy 187 (2017) 601611 605

Start

Initialize the particle

position & velocity

Obtain GRNN output

Calculate Fitness value

Yes Update the particle position &

If (F (i-1) < F (i))
velocity using Eq. (7) & (8).

Store the particle No

Check for maximum No

number of iterations

Yes
Obtain particle with optimal
fitness

Stop

Fig. 3. Flow chart of the proposed methodology.

Fig. 4. Schematic diagram of the experimental setup. 1. Engine, 2. Flywheel, 3. AC dynamometer, 4. Resistive load cell, 5. EGT sensor, 6. Exhaust manifold, 7. Exhaust gas
analyzer, 8. Smoke meter, 9. Pressure transducer, 10. Charge amplifier, 11. Data acquisition system, 12. Computer, 13. Intake plenum, 14. Air heater, 15. Port fuel injector, 16.
Pressure regulator, 17. Fuel pump, 18. Fuel tank, 19. Crank position sensor.

Step 3. Obtain the fitness value for the generated particle using velocity or position shift of the particle in the jth dimension was
the Eq. (10). constrained as follows:
Step 4. Store the fitness value of current particle for further
v z;1 1.
comparison.
v z;2 1.
Step 5. Generate a new particle while updating the position of the
v z;3 1.
current particle as follows:
Obtain the particle with optimal fitness value.

xz;j z 1 xz;j v z;j z 1 9

4. Experimental setup

where z is iteration count or particle generation count and j rep- The experiments were carried on a single cylinder, four-stroke,
resents the dimension of the particle vary from 1 to 3. v z;j is the diesel engine (model TAF1, make: Kirloskar) converted to run on
606 H. Bendu et al. / Applied Energy 187 (2017) 601611

the HCCI mode. The schematic diagram of the experimental test ASTM D6522 standard. The EGR percentage was calculated using
setup is shown in Fig. 4. An electronically programmed port fuel the CO2 ratio between the intake and exhaust gases. The smoke
injector was used to control the quantity of the fuel injection at opacity was measured using an AVL 437 diesel smoke meter [53].
various engine loads at 1500 rpm. The necessary load to the engine The uncertainties of the instruments used are listed in Table 4.
was provided by an electrical dynamometer which was coupled to
the engine. The specifications of the test engine are given in Table 2,
5. Fitness function evaluation
and ethanol was used as the test fuel. Table 3 lists the important
properties of the ethanol. A digital anemometer (make: Lutron,
In this investigation, a fitness function was developed to satisfy
model: AM-4200) was used to measure the intake air during suc-
high brake thermal efficiency (BTE) and low exhaust emissions.
tion. The pressure pulse damping in the intake stroke was mini-
Therefore, the fitness function (F) was directly proportional to
mized by using a surge tank whose volume equal to 60 times of
BTE and inversely proportional to the exhaust emissions (UHC,
the displacement volume. The intake air temperature was varied
CO, NO, and smoke) and EGT.
from 130 to 170 C at a regular interval of 10 C and admitted into 8 9
the test engine. The rate of evaporation of the port injected fuel >
> 1=UHC >>
>
> >
>
essentially depends on the charge temperature. The higher charge >
> >
< 1=CO > =
temperature was desirable for PFI applications [52]. The in-
) F / BTE and F / 1=smoke 10
cylinder pressure was measured using a piezoelectric quartz trans- >
> >
> 1=NO >
> >
>
ducer (make: Kistler, model: 5395A) and a charge amplifier. The >
> >
>
: ;
LabVIEW program was used to capture and average the cylinder 1=EGT
pressure data. The pressure was recorded with a resolution of 0.5
crank angle degree (CAD) interval, and the cycle-to-cycle variations w1  BTE
F
were avoided by averaging a total of 200 consecutive engine cycles. w2  UHC w3  CO w4  smoke w5  NO w6  EGT
The exhaust gas temperature (EGT) was measured with a K-type 11
thermocouple.
The exhaust emissions were sampled using an AVL 444 Digas where w1, w2, w3, w4, w5 and w6 are the weighing parameters cor-
analyzer for measuring CO, CO2, UHC, O2, and NO which follows responds to the BTE, EGT, NO, smoke, CO and UHC respectively.
However, the increase in the BTE leads to an increase in the NO
emission. On the other hand, the decrease in the BTE leads to the
Table 2 increase in the UHC, CO, and smoke emissions. Therefore, it is
Test engine specifications. required to optimize the BTE, while maintaining the lower exhaust
Engine parameters emissions. Therefore, the weights (w1 to w6) are to be decided
based on the emission legislations. The current study deals with
Maximum power 4.4 kW
Rated speed 1500 rpm the specific four cases as follows.
Bore  stroke 87.5 mm  110 mm
Piston type Bowl-in-piston Case 1: Equal priority to all the engine output parameters.
Compression ratio 17.5
Cooling system Air cooling
w1 w2 w3 w4 w5 w6 1=6 12
Case 2: Full priority to exhaust emissions.

Table 3
Important properties of ethanol.
w1 0; w2 0; w3 w4 w5 w6 0:25 13

Chemical and physical properties of ethanol

Case 3: Full priority to performance.
Formula C2H5OH
Molecular weight 46.07
Oxygen content (w %) 34.73 w1 0:7; w2 0:3; w3 w4 w5 w6 0 14
RON 107
Density (at 15 C) (kg/m3) 780 Case 4: High priority to BTE, NO and smoke. Low priority to EGT,
Stoichiometric A/F ratio 9.00 UHC and CO.
Normal boiling point (C) 78
Lower heating value (MJ/kg) 26.9
Latent heat (kJ/kg) 840
w1 0:3; w3 w4 0:2; w2 w5 w6 0:1 15
Boiling point (at 1 atm) (C) 78.4
Note: The detailed explanation for the above four cases are
addressed in the Section 6.5.

Table 4
The accuracy and uncertainty of the instruments used.

Instrument used Range Accuracy % uncertainties

Gas analyzer CO 010% 0.03% 1
UHC 020,000 ppm 10 ppm 0.5
NO 05000 ppm 50 ppm 1
Smoke meter 0100% opacity 1 1
EGT indicator (K-type chromel alumel thermocouples) 0900 C 1 C 0.15
Pressure transducer 0110 bar 0.1 0.1
Speed sensor 010,000 rpm 10 rpm 1
Crank angle encoder 0720 CA 0.5 0.01
Charge amplifier 1 0.1
Load indicator 2506000 W 10 0.2
 H. Bendu et al. / Applied Energy 187 (2017) 601611 607

6. Results and discussion 130 C 140 C 150 C 160 C 170 C

Exhaust Gas Temperature (C)

350
6.1. Engine performance analysis 300
250
The HCCI engine performance was evaluated based on its BTE 200
and EGT. The BTE simultaneously considers two important engine
150
parameters; (i) fuel consumption, and (ii) power output. The EGT
100
specifies the quantity of input thermal energy from the fuel that
has been carried out by exhaust gases. 50
Fig. 5 shows the variation of BTE with engine load for the etha- 0
20 30 40 50 60 70 80 90 100
nol fuelled HCCI operation. The BTE increases with the charge tem-
Engine Load (%)
perature due to the advanced SOC due to high charge temperature.
For PFI applications, the advanced SOC leads to an overall increase Fig. 6. Variation of EGT with engine load.
in the peak in-cylinder gas temperature. A similar trend for an
increase in the BTE is also observed with the engine load. At
170 C charge temperature, a maximum of 45% BTE was found 130 C 140 C 150 C 160 C 170 C
for the ethanol HCCI operation. The BTE increases with the increase 60

UHC Emission (ppm)

in the charge temperature and engine load probably due to the
higher combustion efficiency [51].
Fig. 6 depicts the trend of EGT with engine load for the ethanol
fuelled HCCI operation. It is noted from the figure that, the EGT
increases with the entire load spectrum. At 130 C charge temper-
ature, a maximum of 318 C EGT is noticed. However, the decrease
in EGT is observed with an increase in the charge temperature is
noticed. A similar trend of decrease in EGT with charge tempera-
ture is also observed by Saxena et al. [54]. As the charge tempera-
ture increases then the SOC also advances due to the increase in
the chemical kinetics and rate of reactions during combustion.
Also, the heat transfer losses from the cylinder also increase for
the advanced SOC due to an increase in the residence time of
burned hot gasses.
6.2. Engine emission analysis
Fig. 7 shows the variation of UHC emission with engine load for
the ethanol fuelled HCCI operation. The UHC emission is the indi-
cation of the rate of unburned fuel. As the load increases, the equiv-
alence ratio of the air/fuel mixture also increases which causes a
decrease in the UHC emissions. A maximum UHC emission of
54 ppm is found for 130 C charge temperature and 20% engine
load operation. At 170 C charge temperature, the lowest UHC
emission of 15 ppm is observed during full load operation. The
decrease of UHC emission in the range from 37% to 53% is observed
from 20% load to full load operation in this study.
Fig. 8 shows the variation of CO emission with engine load for
the ethanol fuelled HCCI operation. The CO emission indicates
the rate of incomplete combustion. As the engine load increases,
the peak in-cylinder gas temperature also increases that causes a
reduction in the CO emission. As the charge temperature increases,
the decrease in the CO emission is observed due to the advanced
50
40
30
20
10
0
20 30 40 50 60 70 80 90 100
Engine Load (%)
Fig. 7. Variation of UHC emission with engine load.
SOC with an increase in the charge temperature. The maximum
CO emission of 0.08% is exhibited for 130 C charge temperature
and 20% load operating condition. Similarly, the lowest CO emis-
sion of 0.01% is found at full load and 170 C charge temperature.
The decrease of CO emission in the range from 56% to 71% is
noticed from 20% load to full load operation in this study.
The nitric oxide emissions are produced during the process of
combustion only in the presence of nitrogen and oxygen at
1800 K [55]. Fig. 9 shows the variation of NO emission with engine
load for the ethanol fuelled HCCI operation. The lowest NO emis-
sion is found to be 3 ppm at 20% load and 130 C charge tempera-
ture. Whereas, the highest NO emission of 10 ppm is observed for
170 C charge temperature at full load condition. The increase in
the engine load also causes an increase in the NO emission. This
may be probably due to increase in the peak in-cylinder tempera-
ture [56]. The early SOC also causes the increase in the burned
gases residence time might also increase in the NO emission.
Fig. 10 shows the variation of smoke opacity with engine load for
the ethanol fuelled HCCI operation. The formation of smoke occurs
at extreme air/oxygen deficiency [28]. But, in the port injected HCCI
engines, the utilization of the lean homogeneous air/fuel mixture
causes negligible smoke due to the oxygen availability. As the

130 C 140 C 150 C 160 C 170 C

Brake Thermal Efficiency (%)

50
130 C 140 C 150 C 160 C 170 C
0.09
CO emission (% V/V)

40 0.08
0.07
0.06
30 0.05
0.04
0.03
20
0.02
0.01
10 0
20 30 40 50 60 70 80 90 100 20 30 40 50 60 70 80 90 100
Engine Load (%) Engine Load (%)

Fig. 5. Variation of BTE with engine load. Fig. 8. Variation of CO emission with engine load.
608 H. Bendu et al. / Applied Energy 187 (2017) 601611

130 C 140 C 150 C 160 C 170 C 130 C 140 C 150 C

Nitric Oxide emission (ppm)

12 160 C 170 C
90
10
80

Cylinder pressure (bar)

8
70
6
60
4 50
2 40

0 30
20 30 40
Fig. 9. Variation of NO emission with engine load.
charge temperature increases, the decrease in the smoke opacity is
observed in the entire load spectrum. It can be observed from the
figure that the smoke emissions are almost negligible for the etha-
nol HCCI operation at full load. The increase in the mixture homo-
geneity of the port injected ethanol fuel might be the reason for
lower smoke opacity. At 170 C, the lowest smoke opacity of 0.1%
is observed at full load condition due to the increase in oxidation
temperature. The decrease of smoke emission in the range from
67% to 70% is observed from 20% load to full load operation. The
smoke opacity is below 0.6% throughout this investigation.
6.3. Cylinder pressure crank angle history
The performance and emission parameters of the HCCI engine
mainly depends on the in-cylinder pressure history. Therefore,
the thermodynamics of the cylinder pressure is useful to analyze
the output parameters. Fig. 11 depicts the pressure crank angle
diagram at full load for the ethanol fuelled HCCI operation. It can
be observed that, the increase in the charge temperature causes
advancing the SOC as expected. The advanced SOC and increased
peak pressure are observed with the temperature increment. This
is due to the high reaction rates with an increase in the charge tem-
perature. For high charge temperatures, the fuel/air mixture
reaches the auto-ignition temperature in early crank angles. The
cylinder pressure data is high for 170 C charge temperature at full
load. The peak cylinder pressure is found to be 86 bar for 170 C
charge temperature. The high charge temperature causes the
increase in the cylinder pressure that contributed to the high ther-
mal efficiency and low exhaust emissions. Also, the high charge
temperature results in the decrease of UHC, CO, and smoke emis-
sions, and increase in the NO emission.
6.4. GRNN trainee using cross validation
In this study, Matlab R2015b was used to carry out the simula-
tion by HP desktop with Intel Core i7-3770 3.4 GHz CPU and 8 GB
50 60 70 80 90 100
20
Engine Load (%)
10
0
-30 -20 -10 0 10 20 30
CAD
Fig. 11. Variation of cylinder pressure with crank angle.
RAM. First, GRNN model was trained to complete the prediction
simulation, in which newgrnn() was applied by Matlab to con-
struct a GRNN without training error. The optimal smoothing
parameter r was chosen from the experimental trainee datasets
with a minimum cross-validation error.
The smoothing parameter r was the only adaptive parameter in
the Eq. (1) of the GRNN model. High r values lead to the data over
smoothening, while low r values lead to the data over-fitting.
Hence, optimum r needs to be chosen for effective prediction of
the GRNN model. The optimal r was found from grid search
method in this investigation. The r values [rlow, rhigh] were
defined in a range with N number of iterations. Thus, the cross-
validation was implemented with the obtained N values of Kernel
width was performed using Eq. (16).
rhigh  rlow
r rlow i  1 i 1; 2; . . . ; M 16
N
The optimal r was chosen based on the lowest cross-validation
error of the model. The total number of iterations and the interval
both needs to be constrained for the minimum error (i.e., the effec-
tive search interval was much smaller). In this study, the initial
search interval for the optimal r value was considered as [1; 10],
and search iterations was equal to 10. Fig. 12 shows the influence
of smoothening parameter on GRNN model. The curve has a well-
defined minima, with an optimal r at 4.42.
The developed GRNN tool was tested for 10 number of data
points, and obtained results were compared with the experimental
data sets.
Fig. 13 shows the graphical representation of the percentage of
error during GRNN testing. From the figure it is evident that, a
maximum of 1.5% error is noticed during the testing of the GRNN
network. The proposed GRNN approach can predict the HCCI
engine performance and emission parameters with an accuracy
350

130 C 140 C 150 C 160 C 170 C 300

0.7
CV Error
Smoke Opacity (%)

0.6 250
0.5 4.42
0.4 200
0.3
0.2 150
0.1
0
100
20 30 40 50 60 70 80 90 100 0 2 4 6 8 10 12

Engine Load (%) Sigma

Fig. 10. Variation of smoke opacity with engine load. Fig. 12. Cross-validation curve for GRNN model. Optimal r = 4.42.
 H. Bendu et al. / Applied Energy 187 (2017) 601611 609

of 1.5% for the given input parameters of engine load, charge tem-

Best fitness value of the swarm

0.2
perature and EGR.
0.18

6.5. Hybrid GRNN-PSO optimization 0.16

0.14
The objective of the proposed PSOGRNN algorithm in this
study was to find the optimum operating conditions of HCCI 0.12
engine based on its best performance while minimizing hazardous
0.1
air pollution. The optimization procedure was initialized with the
solutions generated by random and searches for global optima by 0.08
0 20 40 60 80 100 120 140 160
updating the generations. The developed model had a high flexibil-
ity such that the output weights of the model were user defined. Iteration Process
The optimization can be performed for any user defined values Fig. 14. The global best fitness value in every iteration for case 1.
based on their priority of the output criterion. The termination
accuracy of the iteration process is 106. In this paper, four differ-
ent cases were studied to optimize the HCCI engine based on its
0.5

Best fitness value of the swarm

performance and emission parameters.
0.45
6.5.1. Equal priority to all the engine output parameters
0.4
The optimization was initially performed by considering all the
engine output parameters which were equally given priority. This 0.35
case is the preliminary investigation of any engine study that can
be carried out by keeping equal priority to all the outputs. Fig. 14 0.3
shows the best fitness of the generated swarm on the iteration pro-
0.25
cess for equal priority to the engine output parameters. It can be
observed from the figure that, the optimal solution is found at 0.2
127th iteration with the fitness value of 0.1867. The maximum fit- 0 30 60 90 120 150 180
ness value obtained by the 127th particle having values in three Iteration Process
dimensions of [170 C, 62%, 6%].
Fig. 15. The global best fitness value in every iteration for case 3.

6.5.2. Full priority to exhaust emissions

The increase in the emission legislations with an increase in the 0.55
Best fitness value of the swarm

automobile air pollution demands that the priority was mainly

given to the exhaust emissions. The emissions considered in this 0.5
study were NO, smoke, UHC, and CO. These emissions are the pri- 0.45
mary cause of atmospheric pollution. The nitric oxide emissions in
0.4

NO Smoke UHC CO 0.35

(a)
2
0.3
1.5
0.25
1 0 30 60 90 120 150
Iteration Process
% of Error

0.5
0 Fig. 16. The global best fitness value in every iteration for case 4.
0 2 4 6 8 10
-0.5
-1 the engine exhaust cause acid rains, ground-level ozone, and per-
oxyacetyl nitride (PAN) formation in the photochemical smog.
-1.5
Hence, the NO emission levels must be low as possible. The human
-2 health problems caused by smoke (particulate matter) emissions
Scenario
are respiratory morbidity, deficiencies in pulmonary, and lung can-
EGT BTE cer. The environmental effects like accelerated metal corrosion,
(b)
2 and damage to paints, sculptures, etc. with smoke emissions in
1.5 the atmosphere. Hence, these emissions were minimized to its
greater extent as possible. The emissions of CO and UHC indicate
1
low combustion efficiency of the fuel. This can be achieved by
% of Error

0.5 avoiding misfire, wall wetting, and peak cylinder temperature

0 above 1400 K for oxidation of the fuel. Therefore, CO and UHC
-0.5
0 2 4 6 8 10 emissions were also minimized for an efficient HCCI operation.
This case describes the engine with maximum weights to the
-1 emission rates and minimum weight to the thermal efficiency.
-1.5 Therefore, it is suggested that the researchers and engine manufac-
Scenario
turers should not choose this criterion with a full priority to the
Fig. 13. Graphical representation of error percentage for test data. (a) Performance exhaust emissions. From the fitness function, it is evident that this
parameters and (b) emission parameters.
610 H. Bendu et al. / Applied Energy 187 (2017) 601611
Table 5
Analysis of results obtained by hybrid GRNNPSO.
Case Iteration No. Best output swarm
BTE (%) EGT (C)
1 127 40 184
3 147 45 222
4 129 42.3 200
criterion obtained a fitness value of 0 due to the reason of
neglecting thermal efficiency parameter.
6.5.3. Full priority to performance
The performance of the engine is important due to the involve-
ment of its running cost. The low-performance vehicle not only
consumes more fuel, but also causes huge air pollution compared
to high-performance vehicle. The effectiveness of the engine is
measured with BTE (ratio of power output to fuel input energy).
Hence, high BTE is desired for an efficient HCCI engine operation.
The amount of heat energy going as waste without converting into
engine power/work is measured with exhaust gas temperature.
Thus, lower EGT is desired. Fig. 15 shows the best fitness of the
generated swarm on the iteration process for full priority to the
performance parameters. It can be observed from the figure that,
the optimal solution is found at 147th iteration with the fitness
value of 0.4821. The maximum fitness value obtained by the
147th particle having values in three dimensions of [170 C, 76%,
5%].
6.5.4. High priority to BTE, NO and smoke. Low priority to EGT, UHC
and CO
It is understood that the full priority cannot be given to either
performance and emission parameters. Hence, the highly influenc-
ing parameters will be given high priority corresponding to the
HCCI engine performance and emissions. The increase in the per-
formance of IC engine is measured with its BTE and hence it is
given a high priority. The emissions of NO and smoke having
greater influence as they cause acid rain and public health issues
respectively. Fig. 16 shows the best fitness of the generated swarm
on the iteration process for general criteria. It can be observed from
the figure that the optimal solution is found at 129th iteration with
the fitness value of 0.5348. The maximum fitness value obtained by
the 129th particle having values in three dimensions of [170 C,
72%, 4%].
All the above four cases considered are mutually diverse in nat-
ure, and the algorithm is flexible for any user-defined weights. The
weights of this algorithm are chosen based on the individual
nations regulations and legislations. Table 5 compares the results
obtained with the hybrid GRNNPSO for the four cases considered
in Section 5. Among all the four cases, the most general criteria is
case 4 commonly chosen by engine manufacturers with little
modification in the weights. This algorithm took very less time to
converge its optimum fitness value, and hence it can be used for
a rapid engine prototyping.
6.6. Real time application
The proposed algorithm can be used in real time applications to
improve the thermal efficiency of the HCCI engine with minimal
emissions. The power and speed requirement of the engine varies
based on the road condition. In general, the power produced in
each cylinder of a multi-cylinder engine is almost same. Therefore,
all the cylinders of the engine are operated away from the optimal
engine performance. In this case, if some of the cylinders operated
Time taken (s)
NO (ppm) Smoke (%) UHC (ppm) CO (%V)
8 0.15 22 0.025 0.063
9 0.12 18 0.02 0.073
8 0.13 21 0.023 0.064
at optimum engine operating conditions and the rest cylinders to
compensate the excess power, then the overall performance of
the engine will improve and emission will decrease. This algorithm
also reduces time and costs related to experimental activities.
7. Conclusion
In this study, a hybrid GRNNPSO model was developed to
optimize the ethanol-fuelled HCCI engine based on the output per-
formance and emission parameters. The model fits the complicated
nonlinear relationship among the input and output parameters of
the HCCI engine. A total of 45 steady state experimental data sets
were considered for training the GRNN, and newgrnn() was
applied in MATLAB to train the network without any error. The
GRNN network interpretive of the probability estimate such that,
it can predict the performance and emission parameters of HCCI
engine within the range of input parameters. Since GRNN cannot
optimize the solution, and hence swarm based adaptive mecha-
nism was hybridized. A new fitness function was developed by
considering the six engine output parameters. For the developed
fitness function, constrained optimization criteria were imple-
mented in four cases. The optimum HCCI engine operating condi-
tions for the general criteria were found to be 170 C charge
temperature, 72% engine load, and 4% EGR. This model took about
6075 ms for HCCI engine optimization. The main advantage of the
developed fitness function is that a user can define the weights as
per the emission norms and fuel economy standards.
References
[1] Bendu H, Murugan S. Homogeneous charge compression ignition (HCCI)
combustion: mixture preparation and control strategies in diesel engines.
Renew Sustain Energy Rev 2014;38:73246. http://dx.doi.org/10.1016/j.
rser.2014.07.019.
[2] Jimnez-Espadafor FJ, Torres M, Velez JA, Carvajal E, Becerra JA. Experimental
analysis of low temperature combustion mode with diesel and biodiesel fuels:
a method for reducing NOx and soot emissions. Fuel Process Technol
2012;103:5763. http://dx.doi.org/10.1016/j.fuproc.2011.11.014.
[3] Nathan S Swami, Mallikarjuna JM, Ramesh A. Effects of charge temperature
and exhaust gas re-circulation on combustion and emission characteristics of
an acetylene fuelled HCCI engine. Fuel 2010;89:515821. http://dx.doi.org/
10.1016/j.fuel.2009.08.032.
[4] Zheng M, Han X, Asad U, Wang J. Investigation of butanol-fuelled HCCI
combustion on a high efficiency diesel engine. Energy Convers Manag
2015;98:21524. http://dx.doi.org/10.1016/j.enconman.2015.03.098.
[5] Singh AP, Agarwal AK. Combustion characteristics of diesel HCCI engine: an
experimental investigation using external mixture formation technique. Appl
Energy 2012;99:11625. http://dx.doi.org/10.1016/j.apenergy.2012.03.060.
[6] Ishida M, Yamamoto S, Ueki H, Sakaguchi D. Remarkable improvement of NOx-
PM trade-off in a diesel engine by means of bioethanol and EGR. Energy
2010;35:457281. http://dx.doi.org/10.1016/j.energy.2010.03.039.
[7] Alejos Altamirano CA, Yokoyama L, de Medeiros JL, de Queiroz Fernandes
Arajo O. Ethylic or methylic route to soybean biodiesel? Tracking
environmental answers through life cycle assessment. Appl Energy 2016.
http://dx.doi.org/10.1016/j.apenergy.2016.05.017.
[8] Pedrozo VB, May I, Dalla Nora M, Cairns A, Zhao H. Experimental analysis of
ethanol dual-fuel combustion in a heavy-duty diesel engine: an optimisation
at low load. Appl Energy 2016;165:16682. http://dx.doi.org/10.1016/j.
apenergy.2015.12.052.
[9] Maurya RK, Agarwal AK. Experimental investigations of performance,
combustion and emission characteristics of ethanol and methanol fueled
HCCI engine. Fuel Process Technol 2014;126:3048. http://dx.doi.org/10.1016/
j.fuproc.2014.03.031.
 H. Bendu et al. / Applied Energy 187 (2017) 601611
[10] Bahri B, Aziz AA, Shahbakhti M, Said MF Muhamad. Understanding and
detecting misfire in an HCCI engine fuelled with ethanol. Appl Energy
2013;108:2433. http://dx.doi.org/10.1016/j.apenergy.2013.03.004.
[11] Viggiano A, Magi V. A comprehensive investigation on the emissions of ethanol
HCCI engines. Appl Energy 2012;93:27787. http://dx.doi.org/10.1016/j.
apenergy.2011.12.063.
[12] Sjberg M, Dec JE. Effects of EGR and its constituents on HCCI autoignition of
ethanol. Proc Combust Inst 2011;33:30318. http://dx.doi.org/10.1016/j.
proci.2010.06.043.
[13] Huang H, Wang Q, Shi C, Liu Q, Zhou C. Comparative study of effects of pilot
injection and fuel properties on low temperature combustion in diesel engine
under a medium EGR rate. Appl Energy 2016;179:1194208. http://dx.doi.org/
10.1016/j.apenergy.2016.07.093.
[14] Saxena S, Vuilleumier D, Kozarac D, Krieck M, Dibble R, Aceves S. Optimal
operating conditions for wet ethanol in a HCCI engine using exhaust gas heat
recovery. Appl Energy 2014;116:26977. http://dx.doi.org/10.1016/j.
apenergy.2013.11.033.
[15] Maurya RK, Agarwal AK. Experimental study of combustion and emission
characteristics of ethanol fuelled port injected homogeneous charge
compression ignition (HCCI) combustion engine. Appl Energy
2011;88:116980. http://dx.doi.org/10.1016/j.apenergy.2010.09.015.
[16] De Bellis V. Performance optimization of a spark-ignition turbocharged VVA
engine under knock limited operation. Appl Energy 2016;164:16274. http://
dx.doi.org/10.1016/j.apenergy.2015.11.097.
[17] zener O, Yksek L, zkan M. Artificial neural network approach to predicting
engine-out emissions and performance parameters of a turbo charged diesel
engine. Therm Sci 2013;17:15366. http://dx.doi.org/10.2298/
TSCI120321220O.
[18] Oguz H, Sartas I, Baydan HE. Prediction of diesel engine performance using
biofuels with artificial neural network. Expert Syst Appl 2010;37:657986.
http://dx.doi.org/10.1016/j.eswa.2010.02.128.
[19] Deh Kiani MK, Ghobadian B, Tavakoli T, Nikbakht AM, Najafi G. Application of
artificial neural networks for the prediction of performance and exhaust
emissions in SI engine using ethanol-gasoline blends. Energy 2010;35:659.
http://dx.doi.org/10.1016/j.energy.2009.08.034.
[20] Najafi G, Ghobadian B, Tavakoli T, Buttsworth DR, Yusaf TF, Faizollahnejad M.
Performance and exhaust emissions of a gasoline engine with ethanol blended
gasoline fuels using artificial neural network. Appl Energy 2009;86:6309.
http://dx.doi.org/10.1016/j.apenergy.2008.09.017.
[21] Yusaf TF, Buttsworth DR, Saleh KH, Yousif BF. CNG-diesel engine performance
and exhaust emission analysis with the aid of artificial neural network. Appl
Energy 2010;87:16619. http://dx.doi.org/10.1016/j.apenergy.2009.10.009.
[22] Roy S, Banerjee R, Bose PK. Performance and exhaust emissions prediction of a
CRDI assisted single cylinder diesel engine coupled with EGR using artificial
neural network. Appl Energy 2014;119:33040. http://dx.doi.org/10.1016/j.
apenergy.2014.01.044.
[23] Soufi MD, Ghobadian B, Najafi G, Sabzimaleki M, Jaliliantabar F. Performance
and Exhaust Emissions of a SI Two-stroke Engine with Biolubricants Using
Artificial Neural Network. Energy Procedia 2015;75:39. http://dx.doi.org/
10.1016/j.egypro.2015.07.127.
[24] Nikpey H, Assadi M, Breuhaus P, Mrkved PT. Experimental evaluation and
ANN modeling of a recuperative micro gas turbine burning mixtures of natural
gas and biogas. Appl Energy 2014;117:3041. http://dx.doi.org/10.1016/j.
apenergy.2013.11.074.
[25] Mohamed Ismail H, Ng HK, Queck CW, Gan S. Artificial neural networks
modelling of engine-out responses for a light-duty diesel engine fuelled with
biodiesel blends. Appl Energy 2012;92:76977. http://dx.doi.org/10.1016/j.
apenergy.2011.08.027.
[26] Rezaei J, Shahbakhti M, Bahri B, Aziz AA. Performance prediction of HCCI
engines with oxygenated fuels using artificial neural networks. Appl Energy
2015;138:46073. http://dx.doi.org/10.1016/j.apenergy.2014.10.088.
[27] Kara Togun N, Baysec S. Prediction of torque and specific fuel consumption of a
gasoline engine by using artificial neural networks. Appl Energy
2010;87:34955. http://dx.doi.org/10.1016/j.apenergy.2009.08.016.
[28] Srinivasa Pai P, Shrinivasa Rao BR. Artificial Neural Network based prediction
of performance and emission characteristics of a variable compression ratio CI
engine using WCO as a biodiesel at different injection timings. Appl Energy
2011;88:234454. http://dx.doi.org/10.1016/j.apenergy.2010.12.030.
[29] Zurita GV, Pea JC. Vibration based reconstruction of the cylinder pressure in
diesel engines by using neural networks. Investig Y Desarro 2005;60:5360.
[30] Choi Y, Chen J-Y. Fast prediction of start-of-combustion in HCCI with combined
artificial neural networks and ignition delay model. Proc Combust Inst
2005;30:27118. http://dx.doi.org/10.1016/j.proci.2004.08.143.
[31] Kennedy J, Eberhart R. Particle swarm optimization. Proc ICNN95 - int conf
neural networks, vol. 4. IEEE; 1995. p. 19428. http://dx.doi.org/10.1109/
ICNN.1995.488968.
[32] Zhang Q, Ogren RM, Kong S-C. A comparative study of biodiesel engine
performance optimization using enhanced hybrid PSOGA and basic GA. Appl
Energy 2016;165:67684. http://dx.doi.org/10.1016/j.apenergy.2015.12.044.
[33] Manbachi M, Farhangi H, Palizban A, Arzanpour S. Smart grid adaptive energy
conservation and optimization engine utilizing Particle Swarm Optimization
611
and Fuzzification. Appl Energy 2016;174:6979. http://dx.doi.org/10.1016/j.
apenergy.2016.04.083.
[34] Lotfan S, Ghiasi RA, Fallah M, Sadeghi MH. ANN-based modeling and reducing
dual-fuel engines challenging emissions by multi-objective evolutionary
algorithm NSGA-II. Appl Energy 2016;175:919. http://dx.doi.org/10.1016/j.
apenergy.2016.04.099.
[35] Mohammadhassani J, Dadvand A, Khalilarya S, Solimanpur M. Prediction and
reduction of diesel engine emissions using a combined ANNACO method.
Appl Soft Comput J 2015;34:13950. http://dx.doi.org/10.1016/j.
asoc.2015.04.059.
[36] Mousa AA, El-Shorbagy MA, Abd-El-Wahed WF. Local search based hybrid
particle swarm optimization algorithm for multiobjective optimization.
Swarm Evol Comput 2012;3:114. http://dx.doi.org/10.1016/j.
swevo.2011.11.005.
[37] Mozaffari A, Azad NL. An ensemble neuro-fuzzy radial basis network with self-
adaptive swarm based supervisor and negative correlation for modeling
automotive engine coldstart hydrocarbon emissions: a soft solution to a
crucial automotive problem. Appl Soft Comput 2015;32:44967. http://dx.doi.
org/10.1016/j.asoc.2015.04.009.
[38] Roy S, Das AK, Bose PK, Banerjee R. ANN metamodel assisted Particle Swarm
Optimization of the performance-emission trade-off characteristics of a single
cylinder CRDI engine under CNG dual-fuel operation. J Nat Gas Sci Eng
2014;21:115662. http://dx.doi.org/10.1016/j.jngse.2014.11.013.
[39] Park J, Choi J. Optimization of dual-loop exhaust gas recirculation splitting for
a light-duty diesel engine with model-based control. Appl Energy
2016;181:26877. http://dx.doi.org/10.1016/j.apenergy.2016.07.128.
[40] Wong KI, Wong PK, Cheung CS, Vong CM. Modeling and optimization of
biodiesel engine performance using advanced machine learning methods.
Energy 2013;55:51928. http://dx.doi.org/10.1016/j.energy.2013.03.057.
[41] Bertram AM, Zhang Q, Kong S-C. A novel particle swarm and genetic algorithm
hybrid method for diesel engine performance optimization. Int J Engine Res
2016;17:73247. http://dx.doi.org/10.1177/1468087415611031.
[42] Sadeghierad M, Darabi A, Lesani H, Monsef H. Optimal design of the generator
of microturbine using genetic algorithm and PSO. Int J Electr Power Energy
Syst 2010;32:8048. http://dx.doi.org/10.1016/j.ijepes.2010.01.017.
[43] Jeong S, Obayashi S, Minemura Y. Application of hybrid evolutionary
algorithms to low exhaust emission diesel engine design. Eng Optim
2008;40:116. http://dx.doi.org/10.1080/03052150701561155.
[44] You ZP, Zhang WH, Ye XP. Diesel engine inlet and exhaust system fault
detection based on PSO-BP NN. Appl Mech Mater 2014;599601:91821.
http://dx.doi.org/10.4028/www.scientific.net/AMM.599-601.918.
[45] Cai X, Wunsch DC. Engine data classification with simultaneous recurrent
network using a hybrid PSO-EA algorithm. Proc Int Jt Conf Neural Networks
2005;4:231923. http://dx.doi.org/10.1109/IJCNN.2005.1556263.
[46] Janakiraman VM, Nguyen X, Assanis D. Nonlinear identification of a gasoline
HCCI engine using neural networks coupled with principal component
analysis. Appl Soft Comput 2013;13:237589. http://dx.doi.org/10.1016/j.
asoc.2013.01.006.
[47] Nawi NM, Atomi WH, Rehman MZ. The effect of data pre-processing on
optimized training of artificial neural networks. Procedia Technol
2013;11:329. http://dx.doi.org/10.1016/j.protcy.2013.12.159.
[48] Sun Y, Lang M, Wang D, Liu L. A PSO-GRNN model for railway freight volume
prediction: empirical study from China. J Ind Eng Manag 2014;7:41333.
http://dx.doi.org/10.3926/jiem.1007.
[49] Ramsami P, Oree V. A hybrid method for forecasting the energy output of
photovoltaic systems. Energy Convers Manag 2015;95:40613. http://dx.doi.
org/10.1016/j.enconman.2015.02.052.
[50] Ibric S, Djuri J, Parojcic J, Djuric Z. Artificial neural networks in evaluation and
optimization of modified release solid dosage forms. Pharmaceutics
2012;4:53150. http://dx.doi.org/10.3390/pharmaceutics4040531.
[51] Bendu H, Deepak BBVL, Murugan S. Application of GRNN for the prediction of
performance and exhaust emissions in HCCI engine using ethanol. Energy
Convers Manag 2016;122:16573. http://dx.doi.org/10.1016/j.
enconman.2016.05.061.
[52] 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. http://dx.doi.org/10.1016/j.apenergy.2015.01.057.
[53] Kumar Bose P, Banerjee R. An experimental investigation on the role of
hydrogen in the emission reduction and performance trade-off studies in an
existing diesel engine operating in dual fuel mode under exhaust gas
recirculation. J Energy Resour Technol 2012;134:12601. http://dx.doi.org/
10.1115/1.4005246.
[54] Saxena S, Schneider S, Aceves S, Dibble R. Wet ethanol in HCCI engines with
exhaust heat recovery to improve the energy balance of ethanol fuels. Appl
Energy 2012;98:44857. http://dx.doi.org/10.1016/j.apenergy.2012. 04.007.
[55] Habib MA, Elshafei M, Dajani M. Influence of combustion parameters on NOx
production in an industrial boiler. Comput Fluids 2008;37:1223. http://dx.
doi.org/10.1016/j.compfluid.2007.04.006.
[56] Sharma A, Murugan S. Investigation on the behaviour of a DI diesel engine
fueled with Jatropha Methyl Ester (JME) and Tyre Pyrolysis Oil (TPO) blends.
Fuel 2013;108:699708. http://dx.doi.org/10.1016/j.fuel.2012.12.042.