Hindawi

Journal of Sensors
Volume 2017, Article ID 9640546, 10 pages
https://doi.org/10.1155/2017/9640546

Research Article
Power Cell SOC Modelling for Intelligent
Virtual Sensor Implementation

José-Luis Casteleiro-Roca,1 Esteban Jove,1 Fernando Sánchez-Lasheras,2

Juan-Albino Méndez-Pérez,3 José-Luis Calvo-Rolle,1 and Francisco Javier de Cos Juez4
1
Department of Industrial Engineering, University of A Coruña, A Coruña, Spain
2
Department of Construction and Manufacturing Engineering, University of Oviedo, Oviedo, Spain
3
Department of Computer Science and System Engineering, University of La Laguna, Tenerife, Spain
4
Department of Mining Exploitation and Prospecting, University of Oviedo, Oviedo, Spain

Correspondence should be addressed to Fernando Sánchez-Lasheras; [email protected]

Received 8 February 2017; Revised 15 June 2017; Accepted 10 July 2017; Published 28 August 2017

Academic Editor: Eduard Llobet

Copyright © 2017 José-Luis Casteleiro-Roca et al. This is an open access article distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is
properly cited.

Batteries are one of the principal components in electric vehicles and mobile electronic devices. They operate based on
electrochemical reactions, which are exhaustively tested to check their behavior and to determine their characteristics at each
working point. One remarkable issue of batteries is their complex behavior. The power cell type under analysis in this research
is a LFP (Lithium Iron Phosphate LiFePO4). The purpose of this research is to predict the power cell State of Charge (SOC) by
creating a hybrid intelligent model. All the operating points measured from a real system during a capacity confirmation test make
up the dataset used to obtain the model. This dataset is clustered to obtain different behavior groups, which are used to develop the
final model. Different regression techniques such as polynomial regression, support vector regression (SVR), and artificial neural
networks (ANN) have been implemented for each cluster. A combination of these methods is performed to achieve an intelligent
model. The SOC of the power cell can be predicted by this hybrid intelligent model, and good results are achieved.

1. Introduction demands higher autonomy, less charging time, or less weight

at the energy storage components [3].
An electrical energy storage system is the most frequently On the other hand, the electric vehicle forces the devel-
used technology to solve the problem of intermittent energy opment and improvement of storage systems. Currently, a
production in renewable energy systems [1]. In renewable noticeable problem of the electric vehicle is the driving range,
energy systems, the electric energy storage is used as a which is related to the storage system. Although the internal
method of storing the energy when production is higher than combustion engine is inefficient, it does at least provide long
demand and returning the energy when demand is higher driving range due to the high energy density of fossil fuels [4].
than production. Also, these systems are used to supply This paper is focused on battery storage systems and
portable devices or replace the fossil fuels in vehicles, such especially on a Lithium-Ion (Li-Ion) power cell, which is
as in electric vehicles [1]. commonly utilized in electric vehicles. The frequent use of
Nowadays, an increasing number of studies aim to this type of cells makes it possible to identify different charac-
improve electrical energy storage systems. Some research is teristics, such as ageing forecasts, charging, and discharging
focused on dealing with intermittent energy generation in behavior. This information is very important to model the
renewable energy such as solar and wind power systems, power cells properly [5].
proposing to implement those methods in Smart Grid sys- Li-Ion power cells are good candidates for electromobility
tems [2]. Furthermore, the increasing use of portable devices and the renewable energy industry. The reasons why this
like tablets or smartphones, which improve quality of life, battery is suitable for electric drive vehicles are its high
2 Journal of Sensors

power density, high voltage of the cell, low self-discharge,

no memory effect, and long-life cycle [6]. Also, the Li-Ion V
batteries can be found in different chemistries and each one A
+ −
has its specific characteristics, such as energy cells or power T1
cells, fitting the different requirements of the vehicle industry Battery
[7]. i(t)
T2
Due to the extensive use of these batteries in portable
devices as well as in electric vehicles and the need for safety,
the failure analysis and modelling of these batteries are very Figure 1: Capacity confirmation test diagram.
important [8].
Multiple Regression Analysis (MRA) methods are based
on the traditional regression models [9]. These methods are
used in several applications in different subjects due to their
2. Case Study
versatility [10–12]. References [10, 13] show a cost forecast in This research presents the SOC model of a LFP power
the early stage of projects, while [11, 14] evaluate the suppliers cell used in electric vehicle battery systems. This power
performance. The MRA methods are limited in some cases, cell was tested as shown in Figure 1, obtaining different
like in [9] and in [15]. In these articles, MRA techniques were measurements.
used, but due to the nonlinearity of the systems, a different The test carried out and power cell are explained in the
approach was used. However, regression techniques based on subsections in more detail.
soft computing can solve the problems of nonlinear systems,
as was done in [16, 17], where meta-classifiers were applied 2.1. The Battery. One possible method of storing electric
in a predictive control system with the aim of predicting energy is by means of batteries. These devices store energy
the state of the model. In [18, 19], artificial neural networks within an electrochemical medium, capable of storing and
were combined with multiregression in order to optimize then supplying electrical energy due to electrochemical
overbreak prediction. As shown in [20], failures in wind reactions [1]. Batteries are composed of two electrodes and
turbines can be detected and predicted using intelligent an electrolyte, where a redox reaction happens. During the
model techniques. discharging process, the cathode is reduced while the anode
is oxidized, producing a current flow [1]. During the charging
Although some regression problems are addressed with
process, the reaction occurs in reverse order. The number of
the new methods mentioned above, in several cases it is times that the cycle can be repeated is determined by the use
difficult to obtain accurate models, for various reasons. For of the battery, up to the point at which the capacity decreases
instance, a model based on a highly nonlinear system can to unusable levels [30].
lead to a bad performance. Dividing the initial dataset into The most common batteries used in portable devices are
different groups before carrying out the regression could be a the Lithium-Ion batteries. They are widely used in electronic
complementary solution [21, 22]. With this objective in mind, devices and, more recently, in electric vehicles [31]. The Li-
the clustering algorithm 𝐾-means is commonly used [23– Ion cells are characterized by their high energy density, due to
25]. This method consists of splitting up the entire dataset the high voltage they can produce in comparison with other
into a number of clusters. Each piece of data will belong to battery types, while weighing less [1]. They also show a long-
a particular cluster, depending on its characteristics. After life cycle, low self-discharge, and no memory effect [30, 31].
this step, the regression technique is made over each cluster.
Prior researches like [26–29] solve other physical problems 2.2. Capacity Confirmation Test. The capacity confirmation
by using similar techniques. test, which was performed on the battery, measures the capac-
This study develops an intelligent model that is able to ity of the battery in ampere-hour [32]. The measurement of
predict the value of different system variables during a power the capacity was carried out at constant current. The test
cell test. Several clustering techniques were considered for starts with the battery at full charge, so the SOC is at its
maximum. Then, the battery starts to supply constant current
grouping the data with similar behavior to implement the
until it reaches the discharge voltage, which is determined
model. For each cluster, the three aforementioned regression
by the manufacturer [32]. At this point, the battery rests till
methods were tested. The technique selection follows the
the voltage is recovered, so the charging process can begin.
minimum mean squared error criteria. Once the voltage has recovered to a certain value, the battery
The present document is organized as follows. First, an is again charged to its maximum voltage, also specified by
introduction is described. Then, the case of study explains the manufacturer [32]. These processes are performed with
the test carried out and the process followed for obtaining constant current and the present SOC is calculated at each
the dataset. After the case study, the tested algorithms and instant.
the model approach considered in the study are presented. A battery tester was used for the capacity confirmation
The obtained hybrid model and its configuration are shown test. This device is able to charge and discharge the cell
in the Results. Finally, in the last section, future works and within the test parameters, in this case using a constant
conclusions are presented. current and making the corresponding measurements. The
Journal of Sensors 3

3.8
3.6
3.4

Voltage (V)
3.2
3
2.8
2.6
2.4
2.2
2
2800 3000 3200 3400 3600 3800 4000 4200 4400 4600
1 2 3 4
40
30
Current (A) 20
10
0
−10
−20
−30
−40
2800 3000 3200 3400 3600 3800 4000 4200 4400 4600 4800
Number of samples

Figure 2: Current and voltage measured during one cycle.

measurements made are the battery voltage, the current the energy in ampere-hour. The representation of energy,
delivered or absorbed by the battery, the temperature at two at the top of Figure 3, shows how energy in the battery
different points, and how long the test lasts. increases during the charging process and decreases during
Figure 1 shows the test scheme. This figure also shows the discharging process. The representation of energy is
the different measurement devices used and where they are shown as the SOC of the battery, which ranges from 100%
placed: ammeter (A), voltmeter (V), and temperature sensors with the battery fully charged, up to 0% with the battery
(T1 and T2). Given the fact that the temperature of the Li- completely discharged.
Ion batteries is uniform, the sensors location does not affect The two temperature sensors measure the temperature
theoretically the temperature measured. Two redundant tem- while the test is running. The measurement is made at two
perature sensors are used to detect wrong measurements. different locations. Their values also vary cyclically according
The testing device, which supplies or absorbs the current, to the state of the battery, depending on whether it is
is represented as a current source 𝑖(𝑡). charging, discharging, or resting and also depending on the
During this research, the power cell used was a LiFeBATT battery voltage.
X-1P [33], which is a LiFePO4 cell. It is characterized by a The bottom of Figure 3 shows how the temperatures
nominal capacity of 8 Ah and nominal voltage of 3.3 V. evolve in each operating region.
Four different steps can be distinguished in the test The full dataset used in this research has been obtained
(Figure 2): by testing a power cell, as explained before. The battery state
is analyzed using the different parameters measured. Current
(1) Charge: the initial voltage is 3 V and it is increased up and time are used to calculate the SOC of the battery at every
to 3.65 V due to the constant current flowing from the moment. Furthermore, the temperatures were measured in
tester to the battery. order to detect a battery failure during the test. If values are
(2) Rest after the charging process: there is no current far outside a typical range, a malfunction can be detected. All
flowing. During this step, the battery voltage reaches the data measured throughout the capacity battery test was
3.3 V, which is the nominal value. labelled with the aim of determining the SOC.
(3) Discharge: the 3.3 V achieved in the previous step is
decreased to 2 V due to the constant current flowing
3. Model Approach
from the battery to the tester. The model approach is defined by the scheme shown in
(4) Rest after the discharging: there is no current flowing. Figure 4. Taking into consideration the magnitudes measured
The voltage of the battery goes from 2 V to 3 V. A new during the capacity battery test and how the system behaves,
cycle is ready to be started again. the dataset can be split into several operation ranges. Thus,
different clusters are obtained and regression techniques are
Figure 2 shows the voltage and current variation mea- applied for each one. A global model scheme is presented in
sured in one cycle. During the charging, the current is flowing Figure 4. The two inputs of this model are the voltage and
from the tester to the battery, so it is considered positive. On current, and its output is the SOC. The cluster selector links
the other hand, when the current flows from the battery to the selected model with the output. Each cluster represents
the tester, it is considered negative. the best model obtained. The criterion used to assign input
The energy provided or absorbed during the process can data to a specific cluster is the minimum Euclidean distance
be obtained by integrating the current over time, obtaining between the cluster centroids and the input data.
4 Journal of Sensors

100
80

SOC (%)
60
40
20
0
2800 3000 3200 3400 3600 3800 4000 4200 4400 4600 4800
1 2 3 4
43
42
Temperature (∘ C) 41
40
39
38
37
36
35
2800 3000 3200 3400 3600 3800 4000 4200 4400 4600 4800
Number of samples

Figure 3: Energy and temperatures measured in one cycle.

Current Cluster
linear combination of the initial variables. The aim of this
voltage technique is to derive new uncorrelated variables as a linear
selector
combination of the initial variables. This method may help to
reduce the number of variables that represent the data.
Cluster 1
models Data Clustering: The K-Means Algorithm. Clustering is tech-
nique of unsupervised data grouping used to measure simi-
larity [37, 38]. The main aim of the data clustering technique
Cluster 2 is to organize unlabelled vectors into different groups. Thus,
models all similar data will belong to the same cluster [39]. The 𝐾-
SOC means algorithm is frequently used to divide datasets into
.. clusters. This algorithm organizes the samples by minimizing
. the function shown in (1), which represents the squared error
function.
𝑃
Cluster X 󵄩 󵄩2
𝑒 = ∑ ∑ 󵄩󵄩󵄩𝑥 − 𝑐𝑘 󵄩󵄩󵄩 . (1)
models 𝑘=1 𝑥𝜖𝑄𝑘

Figure 4: Model approach. Each cluster is set around a centroid, so the final cluster-
ing will depend on the amount of clusters and on the initial
position of their centroids. The selection of the number of
clusters is very important and requires an initial study of the
The model is obtained as per the diagram of Figure 5. data composition. This value is extremely hard to estimate.
The figure shows only the data division for testing and The partitional algorithm 𝐾-means performs well when the
training. However, a cross-validation technique (k-fold) was data is near its cluster and if each cluster is hyperspherical in
implemented to process the dataset with the aim of achieving shape and sufficiently separated in hyperspace.
better results. The validation of the k-fold model gives a more
real performance and a better generalization [34]. Neural PCA and Maximum Likelihood Hebbian Learning
(MLHL). As mentioned above, PCA is a statistical compress-
3.1. Techniques Used. In this research, different techniques ing data method. The best linear compression of the data can
were used in order to accomplish the best model. These be achieved in terms of LMSE. Some techniques of artificial
techniques are described below. neural networks (ANN) can be used to carry out PCA, for
example, [40, 41].
3.1.1. Clustering A PCA with a negative feedback network [40] can be
used, according to the following definition. Given 𝑁×1 input
Principal Component Analysis. PCA is created by Pearson [35] matrix 𝑥 and 𝑀×1 output matrix 𝑦, a weight matrix 𝑊𝑖𝑗 links
and Hotelling [36]. This statistical technique describes the each output 𝑖 with the input 𝑗, 𝜂 being a learning rate. At the
variation of multivariate data in terms of variables that are beginning of the training process, the network does not have
not correlated. Uncorrelated variables can be expressed as a activation. The weights from the input layer neurons to the
Journal of Sensors 5

Inputs
(K − 1)/K Training
Modelling
data
Outputs

Data

1/K Test Model

Inputs Predicted output
data obtained
Error

Real outputs

Figure 5: Modelling process.

output layer neurons feed back the input data. The output With a first-degree polynomial, the linear summation
neuron activation value is implemented using a linear sum. performed can be described as shown in (6) (for two inputs).
It can be expressed as shown in If the polynomial degree increases, the model obtained
will become more complex. A second-degree polynomial is
𝑁
shown in (7).
𝑦𝑖 = ∑ 𝑊𝑖𝑗 𝑥𝑗 , ∀𝑖 . (2)
𝑗=1 𝑓 (𝑥) = 𝑏0 + 𝑏1 𝑦1 + 𝑏2 𝑦2 (6)
A feedback of the activation function is performed with
the same weights. Then it is subtracted from the inputs and 𝑓 (𝑥) = 𝑏0 + 𝑏1 𝑦1 + 𝑏2 𝑦2 + 𝑏3 𝑦1 𝑦2 + 𝑏4 𝑦12 + 𝑏5 𝑦22 . (7)
the inhibition is performed:
𝑀 Multilayer Perceptron (MLP). The Multilayer Perceptron can
𝑒𝑗 = 𝑥𝑗 − ∑𝑊𝑖𝑗 𝑦𝑖 , ∀𝑗 . (3) be defined as a feedforward artificial neural network (ANN)
𝑖=1 [44, 45]. Its robustness and relative simplicity make this ANN
Then, simple Hebbian learning is carried out between the the one most frequently used. Despite these advantages, the
outputs and the inputs, as shown in good performance of this technique depends heavily on the
ANN architecture selected. The MLP architecture has three
Δ𝑊𝑖𝑗 = 𝜂𝑒𝑗 𝑦𝑖 . (4) different layers: the input, the output, and one or more hidden
ones.
The network learning stability is a consequence of the
negative feedback. In a similar way to Oja’s Subspace algo- A layer has a selected quantity of neurons, with a specific
rithm [41], this network can find the main components of the activation function. Commonly, the same activation function
input data [40]. Therefore, the weights will find a subspace is used in all neurons of the same layer. Activation functions
basis spanned by the current principal components whereas can be of three different types: linear, step, tan-sigmoid,
they will not find the theses components. MLHL [42] is or log-sigmoid. In MLP, the neurons have weighted links
described following the learning rules shown in the following between layers.
equations: a feedforward step (2), a feedback step (3), and The MPL output can be defined as in the following
finally, a weight modification, which is presented in equation [46].
󵄨 󵄨𝑝−1 𝑗
Δ𝑊𝑖𝑗 = 𝜂𝑦𝑖 sign (𝑒𝑗 ) 󵄨󵄨󵄨󵄨𝑒𝑗 󵄨󵄨󵄨󵄨 . (5)
𝑓𝜗 (𝑥) = 𝜉 + ∑𝑎𝑖 𝜙 (V𝑖𝑇 𝑦 + 𝑏𝑖 ) , (8)
𝑖=1
MLHL [42] is connected to the statistical standard
method Exploratory Projection Pursuit (EPP) method [43]. where 𝑦 = (𝑦(1), . . . , 𝑦(𝑑))𝑇 ∈ R𝑑 is the inputs vector;
𝑗 is the hidden layers quantity; 𝜙 is the transfer function
3.1.2. Regression Techniques bounded; 𝜗 = (𝜉, 𝑎1 , . . . , 𝑎𝑘 , 𝑏1 , . . . , 𝑏𝑘 , V11 , . . . , V𝑘𝑑 ) is the
model parameter vector; V𝑖 = (V𝑖1 , . . . , V𝑖𝑑 )𝑇 ∈ R𝑑 is the
Polynomial Regression. A model obtained with polynomial hidden unit 𝑖 parameter vector.
regression [34] can be described as a linear combination of
basis functions. The number of inputs and the polynomial Least Square Support Vector Regression (LS-SVR). Support
degree will affect the number of basis functions. vector regression (SVR) is an adaptation of the support vector
6 Journal of Sensors

machines (SVM) classification algorithm. The principal aim Table 1: Samples assigned to clusters.
of SVR is to plot the data in a high-dimensional space 𝐹 using
nonlinear mapping. Linear regression is then applied in the Cluster Samples
space 𝐹 [47, 48]. (1) Charge 6619
The Least Square formulation of SVM, known as LS- (2) Rest 1089
SVM, is an approximation of the solution of a system. (3) Discharge 7451
Its behavior is very similar to SVM in terms of general (4) Rest 1211
performance [49, 50]. The LS-SVR is the use of LS-SVM Total 16370
for regression [51, 52]. In LS-SVR, a typical squared loss
function replaces an insensitive loss function. This squared
2D PCA
loss function set up the Lagrangian matrix by solving a linear 1
Karush-Kuhn-Tucker equation:
0.95
0 𝜄𝑛𝑇 𝑎0 0
[ ][ ] = [ ], (9) 0.9
𝜄𝑛 𝐾 + 𝛾−1 𝜄 𝑎 𝑦
where 𝜄𝑛 is the 𝑛 ones vector; 𝑇 is the matrix or vector means 0.85
transpose; 𝛾 is the weight vector; 𝑎 is the regression vector; 𝑎0
is the model offset. 0.8
To implement this technique two parameters are
required: 𝛾 and 𝜎, where 𝜎 is the kernel width [53]. 0.75

3.2. Preprocessing the Dataset. The dataset was obtained 0.7

during the described capacity confirmation test performed
on the power cell. The State of Charge was determined by the 0.65
−1 −0.5 0 0.5 1 1.5
current and the time for each test. Voltage and current were
monitored to determine the SOC. As explained in previous Figure 6: PCA analysis.
sections, temperatures were measured at two points of the
battery in order to find malfunctions. 2D MLHL
The SOC parameter of the battery should go sequentially 6
from 0% to 100% and then from 100% to 0%. All incomplete
cycles which did not fulfill this condition were discarded. 5
After the data filtering, only completed cycles were taken
into consideration for the implementation of the model. The 4
measured data was registered with a sample of 1 Hertz. After
this discard, the dataset for implementing the model was 3
reduced from 18131 to 16370 samples.
The cycle consists of four different states: charge (1), rest 2
(2), discharge (3), and rest (4). During the test, all data
measured was labelled with the current state of the power cell. 1
Before developing the regression models, clustering algo-
rithms were applied to the dataset. For this purpose, an 0
analysis with MHLH and PCA was performed. The analysis
showed that four different groups of samples could be −1
−2 −1.5 −1 −0.5 0 0.5 1 1.5 2 2.5 3
appreciated, as Figures 6 and 7 show. Following the criteria
suggested by PCA and MHLH, 𝐾-means algorithm was Figure 7: MLHL analysis.
applied to create the four clusters. Each group represents the
four different states of the battery during the test.
The number of samples assigned to each cluster is shown applied by 𝐾-means and the clusters accomplished had 100%
in Table 1. It is remarkable that the cluster distribution correct assignation. Due to the good performance achieved,
corresponds to the four steps of the capacity confirmation test the model approach is not required to know the state of the
mentioned in Section 2.2. Polynomial regression, MLP-ANN, cycle.
and LS-SVR were the three techniques trained for the clusters. The MLP-ANN algorithm was implemented with dif-
ferent configurations; the number of hidden layer neurons
4. Results changed from 2 to 15, while only one hidden layer was used.
All hidden layer neurons shared the same activation function
The real state of the cell labelled during the test was com- (tan-sigmoid). A linear activation function was used as an
pared with the data distribution obtained as a result of the output layer neuron. The Levenberg-Marquardt algorithm
clustering. The dataset has the requisite characteristics to be was applied on the training process. Learning algorithm was
Journal of Sensors 7

25 0.1

20 0.095

15 0.09

10 0.085

0.08
5
0.075
0
ANN Poly SVR ANN Poly SVR
(a) Cl 1 (b) Cl 2

20
0.16

15 0.14

0.12
10
0.1
5
0.08

0 0.06
ANN Poly SVR ANN Poly SVR
(c) Cl 3 (d) Cl 4

Figure 8: Boxplot of MSE by cluster.

Table 2: Best MSE achieved for each technique. Table 3: MSE for the best methods.

Cycle state MLP Poly LS-SVR Cycle state Model MSE

(1) Charge 0.1094 23.9967 0.1392 (1) Charge MLP, 10 neurons 0.1392
(2) Rest 0.0921 0.0904 0.0902 (2) Rest MLP, 8 neurons 0.4767
(3) Discharge 0.4767 17.8144 0.5052 (3) Discharge LS-SVR 0.0902
(4) Rest 0.0917 0.0930 0.0865 (4) Rest LS-SVR 0.0865

based on the gradient decrease, and MSE was set as the 4.1. Statistical Analysis. Here, the results of the statistical
criterion. analysis are shown with the aim of determining the accuracy
The LS-SVR regression algorithm was trained with the of the selected models compared with that of those nonse-
autotuning implemented in the Matlab toolbox developed lected.
by KU Leuven-ESAT-SCD. The chosen model kernel was
For each cluster, the best model selected is compared
Radial Basis Function (RBF) and “Function Estimation” the
statistically with the others, in terms of MSE, to determine
kind selected to make the regression. The cost-criterion is
if there is a difference between their MSEs.
“leaveoneoutlssvm,” with an MSE performance function and
an optimization function type “simplex.” Figure 8 shows the boxplot for each cluster using ANN-
The polynomial regression technique was applied with MLP, polynomial regression, and LS-SVR, while the results of
varying polynomial order from 1 to 10. the statistical analysis using a 𝑡-test analysis with a 5% 𝛼-level
The Mean Square Error (MSE) value is used to compare are shown in Table 4.
the accuracy of the models. The lowest MSEs obtained for Based on the results of Table 4, the following may be
each technique are shown in Table 2. concluded:
In Table 3, the best regression results are shown, as well as
the technique applied to achieve them. (i) Cluster 1: there is no statistical difference between
The technique selection takes into consideration the ANN-MLP, 10 neurons, and LS-SVR.
computational cost when the MSEs achieved for two or more (ii) Cluster 2: there is no statistical difference between any
techniques are similar. of the three models.
8 Journal of Sensors

Table 4: 𝑝 value for each model in each cluster.

Cluster Selected model Compared model 𝑝 value

1 ANN-MLP, 10 neurons Polynomial 9.44 ⋅ 10−5
1 ANN-MLP, 10 neurons SVR 0.1364
2 LS-SVR ANN-MLP, 10 neurons 0.0695
2 LS-SVR Polynomial 0.0735
3 ANN-MLP, 8 neurons Polynomial 2.876 ⋅ 10−9
3 ANN-MLP, 8 neurons SVR 0.4789
4 LS-SVR ANN-MLP, 10 neurons 0.4970
4 LS-SVR Polynomial 0.8713

(iii) Cluster 3: there is no statistical difference between [2] H. Qian, J. Zhang, and J.-S. Lai, “A grid-tie battery energy
ANN-MLP, 10 neurons, and LS-SVR. storage system,” in Proceedings of the 2010 IEEE 12th Workshop
on Control and Modeling for Power Electronics, COMPEL 2010,
(iv) Cluster 4: there is no statistical difference between any pp. 1–5, June 2010.
of the three models. [3] N. Chaturvedi, R. Klein, J. Christensen, J. Ahmed, and A. Kojic,
“Modeling, estimation, and control challenges for lithium-ion
In any of the previous cases, the difference between the batteries,” in Proceedings of the American Control Conference
MSEs, in the best case, is lower than 0.0052 so it can be (ACC), pp. 1997–2002, 2010.
considered that the model selected does not significantly [4] S. Vukosavic, Electrical Machines, Power Electronics and Power
affect the results. Systems, Springer, 2012.
[5] O. Konig, S. Jakubek, and G. Prochart, “Battery impedance
5. Conclusions and Future Works emulation for hybrid and electric powertrain testing,” in
Proceedings of the 2012 IEEE Vehicle Power and Propulsion
The approach used in this study presents remarkable results. Conference, VPPC 2012, pp. 627–632, October 2012.
The MSE average is 0.1815, varying from 0.0865 to 0.4767, [6] N. H. F. Ismail and S. F. Toha, “State of charge estimation of a
depending on the variable and the cycle state. Lithium-ion battery for electric vehicle based on particle swarm
The State of Charge value for the confirmation of the optimization,” in Proceedings of the 2013 IEEE International
battery capacity can be predicted in real time. The model Conference on Smart Instrumentation, Measurement and Appli-
obtained is capable of checking that the power cell test is cations, ICSIMA 2013, pp. 1–4, November 2013.
functioning correctly when it offers wrong results. [7] M. Swierczynski, D.-I. Stroe, A.-I. Stan, R. Teodorescu, and S. K.
The overall efficiency of this novel approach is increased Kær, “Lifetime estimation of the nanophosphate lifepo4 battery
chemistry used in fully electric vehicles,” IEEE Transactions on
if a hybrid model is implemented. This way, each local model
Industry Applications, vol. 51, no. 4, pp. 3453–3461, 2015.
was obtained by training one group of the initial dataset.
Better results were achieved by employing MLPs and LS-SVR. [8] J. C. Álvarez Antón, P. J. Garcı́a Nieto, F. J. de Cos Juez, F. S.
Lasheras, C. B. Viejo, and N. Roqueñı́ Gutiérrez, “Battery state-
However, the difference between the two, in terms of MSE, is
of-charge estimator using the MARS technique,” IEEE Transac-
not statistically great, so both models may be used in any of tions on Power Electronics, vol. 28, no. 8, pp. 3798–3805, 2013.
the clusters.
[9] C. C. Turrado, M. D. C. M. López, F. S. Lasheras, B. A. R. Gómez,
The temperature measured in the cell is highly important
J. L. C. Rollé, and F. J. D. C. Juez, “Missing data imputation
parameter relating to health of the device. Using the tem- of solar radiation data under different atmospheric conditions,”
peratures measured in the tests, a study and a modelling of Sensors, vol. 14, no. 11, pp. 20382–20399, 2014.
these parameters will be undertaken to detect deviations in [10] R. Jin, K. Cho, C. Hyun, and M. Son, “MRA-based revised CBR
behavior. This new model would be used with the one-class model for cost prediction in the early stage of construction
classifier algorithm in order to detect any faults in this type of projects,” Expert Systems with Applications, vol. 39, no. 5, pp.
systems. 5214–5222, 2012.
[11] L.-H. Ho, S.-Y. Feng, Y.-C. Lee, and T.-M. Yen, “Using modified
Conflicts of Interest IPA to evaluate supplier’s performance: multiple regression
analysis and DEMATEL approach,” Expert Systems with Appli-
The authors declare that there are no conflicts of interest cations, vol. 39, no. 8, pp. 7102–7109, 2012.
regarding the publication of this paper. [12] J. Osborn, D. Guzman, F. J. D. C. Juez et al., “Open-loop tomog-
raphy with artificial neural networks on CANARY: on-sky
References results,” Monthly Notices of the Royal Astronomical Society, vol.
441, no. 3, pp. 2508–2514, 2014.
[1] C. Chukwuka and K. A. Folly, “Batteries and super-capacitors,” [13] P. J. Garcı́a Nieto, J. R. Alonso Fernández, F. Sánchez Lasheras,
in Proceedings of the IEEE Power and Energy Society Conference F. J. de Cos Juez, and C. Dı́az Muñiz, “A new improved
and Exposition in Africa: Intelligent Grid Integration of Renew- study of cyanotoxins presence from experimental cyanobacteria
able Energy Resources, PowerAfrica 2012, pp. 1–6, July 2012. concentrations in the Trasona reservoir (Northern Spain) using
Journal of Sensors 9

the MARS technique,” Science of the Total Environment, vol. 430, [29] J. R. Alonso Fernández, C. Dı́az Muñiz, P. J. Garcia Nieto, F. J. de
pp. 88–92, 2012. Cos Juez, F. Sánchez Lasheras, and M. N. Roqueñı́, “Forecasting
[14] G. Flórez, P. A. Saiz, P. Garcı́a-Portilla et al., “Predictors the cyanotoxins presence in fresh waters: a new model based
of posttreatment drinking outcomes in patients with alcohol on genetic algorithms combined with the MARS technique,”
dependence,” European Addiction Research, vol. 21, no. 1, pp. 19– Ecological Engineering, vol. 53, pp. 68–78, 2013.
30, 2015. [30] Wikipedia, Lithium-ion battery - wikipedia, the free encyclope-
[15] Y. Cho and H. B. Awbi, “A study of the effect of heat source loca- dia, [Online; accessed 30-January-2014] (2014). http://en.wiki-
tion in a ventilated room using multiple regression analysis,” pedia.org/w/index.php?title=Lithium-ion batteryoldid=592970854.
Building and Environment, vol. 42, no. 5, pp. 2072–2082, 2007. [31] A. R. Sparacino, G. F. Reed, R. J. Kerestes, B. M. Grainger,
[16] J. Nieves, I. Santos, and P. G. Bringas, “Enhancing the prediction and Z. T. Smith, “Survey of battery energy storage systems and
stage of a model predictive control systems through meta- modeling techniques,” in Proceedings of the 2012 IEEE Power and
classifiers,” Dyna, vol. 88, no. 3, pp. 290–298, 2013. Energy Society General Meeting, PES 2012, pp. 1–8, July 2012.
[17] A. S. Sánchez, P. R. Fernández, F. S. Lasheras, F. J. de Cos Juez, [32] PNGV Battery Test Manual (2 2001).
and P. J. Nieto, “Prediction of work-related accidents according [33] LiFeBATT X-1P 8Ah 38123 Cell (3 2011).
to working conditions using support vector machines,” Applied [34] C. M. Bishop, Pattern Recognition and Machine Learning (Infor-
Mathematics and Computation, vol. 218, no. 7, pp. 3539–3552, mation Science and Statistics), Springer, New York, NY, USA,
2011. 2006.
[18] H. Jang and E. Topal, “Optimizing overbreak prediction based [35] K. Pearson, “On lines and planes of closest fit to systems of
on geological parameters comparing multiple regression analy- points in space,” Philosophical Magazine, vol. 2, no. 6, pp. 559–
sis and artificial neural network,” Tunnelling and Underground 572, 1901.
Space Technology, vol. 38, pp. 161–169, 2013. [36] H. Hotelling, “Analysis of a complex of statistical variables into
[19] F. Sánchez Lasheras, F. J. de Cos Juez, A. Suárez Sánchez, A. principal components,” Journal of Educational Psychology, vol.
Krzemień, and P. Riesgo Fernández, “Forecasting the COMEX 24, no. 6, pp. 417–441, 1933.
copper spot price by means of neural networks and ARIMA [37] A. K. Qin and P. N. Suganthan, “Enhanced neural gas network
models,” Resources Policy, vol. 45, pp. 37–43, 2015. for prototype-based clustering,” Pattern Recognition, vol. 38, no.
[20] A. Alvarez-Huerta, R. Gonzalez-Miguelez, D. Garcia-Metola, 8, pp. 1275–1288, 2005.
and A. Noriega-Gonzalez, “Failure detection and prediction in [38] J. Ye and T. Xiong, “SVM versus Least Squares SVM,” Journal of
wind turbines by using scada data,” Dyna, vol. 86, no. 4, pp. 467– Machine Learning Research, vol. 2, pp. 644–651, 2007.
473, 2011.
[39] S. Kaski, J. Sinkkonen, and A. Klami, “Discriminative cluster-
[21] D. Martı́nez-Rego, O. Fontenla-Romero, and A. Alonso- ing,” Neurocomputing, vol. 69, no. 1-3, pp. 18–41, 2005.
Betanzos, “Efficiency of local models ensembles for time series
prediction,” Expert Systems with Applications, vol. 38, no. 6, pp. [40] C. Fyfe, “A neural network for PCA and beyond,” Neural
6884–6894, 2011. Processing Letters, vol. 6, no. 1-2, pp. 33–41, 1997.
[22] A. Suárez Sánchez, F. J. Iglesias-Rodrı́guez, P. Riesgo Fernández, [41] E. Oja, “Neural networks, principal components, and sub-
and F. J. de Cos Juez, “Applying the K-nearest neighbor tech- spaces,” International Journal of Neural Systems, vol. 01, no. 01,
nique to the classification of workers according to their risk pp. 61–68, 1989.
of suffering musculoskeletal disorders,” International Journal of [42] M. Saǧsan, I. T. Medeni, and T. D. Medeni, “Knowledge
Industrial Ergonomics, vol. 52, pp. 92–99, 2014. Management Paradigms: Implementation through individual
[23] A. Cherif, H. Cardot, and R. Boné, “SOM time series clustering fuzzy-based Education,” in Proceedings of the 12th International
and prediction with recurrent neural networks,” Neurocomput- Conference on Application of Fuzzy Systems and Soft Computing,
ing, vol. 74, no. 11, pp. 1936–1944, 2011. ICAFS 2016, pp. 259–266, Vienna, Austria, August 2016.
[43] J. H. Friedman and J. W. Tukey, “A projection pursuit algorithm
[24] M. H. Ghaseminezhad and A. Karami, “A novel self-organizing
for exploratory data analysis,” IEEE Transactions on Computers,
map (SOM) neural network for discrete groups of data cluster-
vol. 23, no. 9, pp. 881–890, 1974.
ing,” Applied Soft Computing Journal, vol. 11, no. 4, pp. 3771–
3778, 2011. [44] P. Wasserman, Advanced Methods in Neural Computing, John
Wiley & Sons, New York, NY, USA, 1st edition, 1993.
[25] J.-L. Casteleiro-Roca, J. L. Calvo-Rolle, J. A. M. Pérez, N. R.
Gutiérrez, and F. J. de Cos Juez, “Hybrid intelligent system to [45] Z. Zeng and J. Wang, Advances in Neural Network Research And
perform fault detection on BIS sensor during surgeries,” Sensors, Applications, Springer Publishing Company, New York City, NY,
vol. 17, no. 1, article 179, 2017. USA, 1st edition, 2010.
[26] M. J. Crespo-Ramos, I. Machón-González, H. López-Garcı́a, [46] J. Rynkiewicz, “General bound of overfitting for MLP regression
and J. L. Calvo-Rolle, “Detection of locally relevant variables models,” Neurocomputing, vol. 90, pp. 106–110, 2012.
using SOM-NG algorithm,” Engineering Applications of Artifi- [47] N. Cristianini and J. Shawe-Taylor, An Introduction to Support
cial Intelligence, vol. 26, no. 8, pp. 1992–2000, 2013. Vector Machines, Cambridge University Press, New York, NY,
[27] J. L. Calvo-Rolle, J. L. Casteleiro-Roca, H. Quintián, and M. Del USA, 2000.
Carmen Meizoso-Lopez, “A hybrid intelligent system for PID [48] I. Steinwart and A. Christmann, Support Vector Machines,
controller using in a steel rolling process,” Expert Systems with Springer Publishing Company, New York, NY, USA, 1st edition,
Applications, vol. 40, no. 13, pp. 5188–5196, 2013. 2008.
[28] I. Machón-González, H. López-Garcı́a, and J. L. Calvo-Rolle, [49] J. A. K. Suykens and J. Vandewalle, “Least squares support
“Neuro-robust controller for non-linear systems,” Dyna, vol. 86, vector machine classifiers,” Neural Processing Letters, vol. 9, no.
no. 3, pp. 308–317, 2011. 3, pp. 293–300, 1999.
10 Journal of Sensors

[50] R. Wang, A. Wang, and Q. Song, “Research on the Alkalinity of

Sintering Process Based on LS-SVM Algorithms,” in Advances
in Computer Science and Information Engineering, vol. 168
of Advances in Intelligent and Soft Computing, pp. 449–454,
Springer, Berlin, Germany, 2012.
[51] Y. Guo, X. Li, G. Bai, and J. Ma, “Time series prediction method
based on LS-SVR with modified gaussian RBF,” in Neural
Information Processing, vol. 7664 of Lecture Notes in Computer
Science, pp. 9–17, Springer, Berlin, Germany, 2012.
[52] L. Wang and J. Wu, “Neural network ensemble model using
PPR and LS-SVR for stock market forecasting,” in Advanced
Intelligent Computing, vol. 6838 of Lecture Notes in Computer
Science, pp. 1–8, Springer, Berlin, Germany, 2012.
[53] Y. Li, X. Shao, and W. Cai, “A consensus least squares support
vector regression (LS-SVR) for analysis of near-infrared spectra
of plant samples,” Talanta, vol. 72, no. 1, pp. 217–222, 2007.
 International Journal of

Rotating
Machinery

International Journal of
The Scientific
(QJLQHHULQJ Distributed
Journal of
Journal of

Hindawi Publishing Corporation

World Journal
Hindawi Publishing Corporation Hindawi Publishing Corporation
Sensors
Hindawi Publishing Corporation
Sensor Networks
Hindawi Publishing Corporation
http://www.hindawi.com Volume 201 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Journal of

Control Science
and Engineering

Advances in
Civil Engineering
Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

Submit your manuscripts at

https://www.hindawi.com

Journal of
Journal of Electrical and Computer
Robotics
Hindawi Publishing Corporation
Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014

VLSI Design
Advances in
OptoElectronics
,QWHUQDWLRQDO-RXUQDORI

International Journal of
Modelling &
Simulation
$HURVSDFH
Hindawi Publishing Corporation Volume 2014
Navigation and
Observation
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
in Engineering
Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014
(QJLQHHULQJ
+LQGDZL3XEOLVKLQJ&RUSRUDWLRQ
KWWSZZZKLQGDZLFRP 9ROXPH
Hindawi Publishing Corporation
http://www.hindawi.com
http://www.hindawi.com Volume 201-

International Journal of
International Journal of Antennas and Active and Passive Advances in
Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration
Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation
http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014