Hindawi

Scientific Programming
Volume 2022, Article ID 8455629, 19 pages
https://doi.org/10.1155/2022/8455629

Research Article
Aircraft Failure Rate Prediction Method Based on CEEMD and
Combined Model

Wenqiang Li and Ning Hou

School of Mechatronics Engineering, Shenyang Aerospace University, Shenyang 110136, China

Correspondence should be addressed to Wenqiang Li; [email protected]

Received 12 March 2022; Revised 28 April 2022; Accepted 25 May 2022; Published 24 June 2022

Academic Editor: Sheng Du

Copyright © 2022 Wenqiang Li and Ning Hou. 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.
Accurate prediction of aircraft failure rate can improve flight safety and spare parts supply efficiency and effectively provide good
maintenance and maintenance decisions and health management guidance. In order to achieve accurate prediction of non-linear
and non-stationary aircraft failure rate, an aircraft failure rate prediction method based on the fusion of complementary ensemble
empirical mode decomposition (CEEMD) and combined model is proposed. Firstly, the complementary set empirical mode is
used to decompose the failure rate into multiple components with different frequencies, then the integrated moving average
autoregressive model (ARIMA) model and grey Verhulst model are selected to predict different components, the entropy weight
method is used to solve the coefficients of the combined model, and finally the prediction results of each prediction model are
multiplied by their respective weight coefficients to obtain the final prediction results. The experiment was carried out by taking
the actual case application of the failure rate data of the aircraft fuel control system as an example. Seven evaluation functions are
used as evaluation criteria to evaluate the performance of the combined model. Experimental results show that the developed
combined model is better than other models such as sum of squared error (SSE) and mean absolute error (MAE), which can
significantly improve the prediction accuracy of aircraft failure rate. It is proved that the model can improve the accuracy and
effectiveness of aircraft failure rate prediction. At the same time, the stability of the model has certain advantages over other
models and has a good application prospect.

1. Introduction factors threatening flight safety. With the rapid development

of aircraft manufacturing industry and the continuous
As one of the core and key technologies in many fields, improvement of aircraft performance, military aviation has
prediction technology has been widely used, such as in the higher and higher requirements for reducing aircraft fail-
fields of geology [1–4], transportation [5], water conservancy ures. Scientific prediction of aircraft failures can make sci-
[6], and wind energy [7–16]. It plays an important sup- entific decisions on aviation maintenance, diagnosis, and
porting role in decision making and management. As a health management. It is an important component of im-
highly integrated and complex system, aircraft may break proving maintenance support capabilities. It can effectively
down at any time in harsh working environment for a long improve aircraft safety, enhance reliability and maintain-
time. In order to prevent flight accidents and even air crashes ability, and reduce maintenance and insurance-related costs.
caused by aircraft failures, people are actively looking for Failure prediction is generally divided into failure time
more efficient aircraft health management and maintenance prediction [17], life prediction [18], and failure rate pre-
decision-making solutions to improve flight safety and re- diction [19]. Aircraft failure rate is an important parameter
liability. According to statistics, aircraft failures in the field of of reliability, maintainability, and supportability of aviation
civil aviation show a downward trend year by year and are equipment. Due to the complexity of aircraft fault cross-
becoming more and more safe. However, there are still some linking, strong uncertainty and randomness of fault causes,
2 Scientific Programming

Table 1: Data-driven prediction method.

Data-driven method failure rate prediction
Single model Combination model
Decomposition
Machine learning model Statistical model Grey model Multi-model combination integration
combination
Holt–Winters Yang et al.
Al-Garni
Neural Yang seasonal model [33] EMD-LS- Wang
et al. ARIMA
network et al. [28] Neural network Wang et al. SVM et al. [42]
[23, 24]
Li et al. regression [34]
GM(1, 1)
[31] Celikmih
Improved ANN-SVR-LR
Hu et al. Li et al. et al. [35] EMD- Wang
neural ARMA
[25] [29] Gu et al. RVM-GM et al. [43]
network IFT
[36]
Zhang et al.
Genetic SVR-MLR-PCA
Cheng et al. [37]
neural
[26] Random forest Yan et al.
network
combination [38]
Weibull Zhang Grey Zhang RVEMD- Xu et al.
Xu et al.
distribution et al. [30] Verhulst et al. [32] ARMA-BP GMDH [44]
[39]
Zhang et al.
SVM Generalized
[27] Liang et al.
weighted least
[40]
squares

few fault samples, and other problems, the modeling and verification were used based on two years of data [24]. The
prediction of aircraft failure rate has always been a research application of artificial neural network (ANN) is relatively
hotspot of many scholars, and the research in this area has successful, but it is still limited by the slow learning speed
great theoretical value and practical significance. and the need for a lot of numerical training, so it is difficult to
In the past few decades, researchers have proposed a obtain satisfactory prediction accuracy. Some scholars use
variety of models and methods for predicting failure rates, improved neural network [25] and genetic neural network
which are mainly divided into two categories, namely, [26] to predict the aircraft failure rate, which has more
model-based failure rate prediction and data-driven method advantages in aircraft failure rate prediction. Because of the
failure rate prediction [20]. Due to the complex composition strictness of the algorithm, the prediction results are more
of the aircraft system and the cross-linking of each system accurate, but the structure of the model is too complex, and
level, unit level, and component level, there are many in- modeling takes more time. At the same time, when the
terferences and mutual influences in the use process, so it is amount of failure rate data is small, the prediction effect of
difficult to accurately model the aircraft failure rate distri- the above model is not ideal. Also, for the failure rate
bution function, and improper modeling will also produce prediction problem, the support vector machine (SVM)
large errors in prediction, thus limiting the effective ap- prediction method [27] is used for prediction. SVM can use
plication of model-based failure rate prediction methods. the kernel function to solve non-linear problems, which is
Therefore, it is difficult to use this method to predict the suitable for small sample prediction, but it is difficult to
aircraft failure rate [21]. The application of data-driven implement for large-scale training samples. Yang et al. [28]
prediction methods is more flexible. The main advantage is proposed a seasonal-integrated moving average autore-
that the trained model data can be easily updated with new gressive (ARIMA) model and analyzed and compared the
and updated data. Therefore, data-driven failure rate pre- forecast results. Seasonally integrated moving average
diction has been widely concerned and applied [22]. autoregressive SARIMA (0, 1, 1) (0, 1, 1)12 model is suitable
Data-driven aircraft failure rate prediction models can for aircraft failure rate prediction, which verifies the feasi-
be divided into single-item models and combined models. bility and effectiveness of the model. Li and Kang [29] used
The related literature is shown in Table 1. Single-item models the autoregressive moving average (ARMA) model to pre-
usually include machine learning models, statistical models, dict the failure rate. Zhang et al. [30] proposed the gener-
and grey models. Al-Garni et al. [23] established an artificial alized regression equation based on Weibull distribution to
neural network model for predicting the failure rate of Dash- predict the failure of aircraft auxiliary power unit and
8 aircraft tires, used six years of historical data for model verified the performance of the proposed model using the
construction and verification, and applied the artificial three-year dataset provided by China Southern Airlines.
neural network model to verify the applicability of future tire When the statistical model encounters large changes in the
failure rate prediction. At the same time, the artificial neural outside world, it often has large deviations, resulting in low
network model was applied in the reliability analysis of accuracy. The ARIMA model has effectively expanded from
Boeing 737 tires, and the failure rate prediction model and stationary time series to non-stationary cases. It is an
Scientific Programming 3

effective integration of autoregressive, difference, and airborne equipment failure rate in different working envi-
moving average models. It is a better time series statistical ronments and achieved certain prediction results. Zhang
model and has been widely used. Li et al. [31] aimed at the et al. [37] comprehensively used a variety of data analysis
problem of high failure rate of Cessna 172 basic aircraft in techniques such as support vector regression (SVR), mul-
operation, and a GM (1,1) fault prediction method based on tiple regression (MLR), and principal component analysis
grey theory was proposed. This method can effectively (PCA) and proposed a comprehensive prediction method.
predict the failure of Cessna 172 basic aircraft and realize the The mathematical relationship between the influencing
prediction of aircraft failure. However, this method some- factors and failure rate was studied when the failure rate was
times has large deviation and failure. Also, the failure rate is collected from 2000 to 2003, and the prediction results verify
predicted by using the non-equidistant grey Verhulst the effectiveness of this method. At the same time, many
modified model [32]. However, the grey prediction is not researchers have also proposed some other combination
suitable for a large number of data and cannot be used for forecasting methods, including random forest combination
long-term prediction. The grey Verhulst model can solve the [38], ARMA-BP combination [39], generalized weighted
problems of less historical data, low sequence integrity, and least squares combination [40], optimal combination [41],
reliability. It can generate irregular original data to obtain etc. The advantage of the multi-model combination method
strong sequence. It is suitable for non-monotonic swing is that it combines the advantages of various prediction
sequence. Compared with other grey models, it has certain models and reduces the prediction error, so it obtains better
advantages and can improve the overall prediction accuracy. prediction accuracy and prediction effect. Although it
Although the above single prediction model or method has combines the advantages of other single prediction models,
achieved good prediction results, for the aircraft system, due it is difficult to optimize the single model involved in the
to the complexity of its system composition and failure combined model and determine the parameters such as the
mechanism, there is a strong coupling between different combined weight coefficient. At the same time, the com-
components, and the single failure prediction method has its bined prediction model is too complex and has limitations in
own advantages and disadvantages. Due to its own limita- dealing with the non-linearity and instability of aircraft
tions and applicable conditions, the use of a single model for failure rate, which needs to be deeply discussed in practical
fault prediction has corresponding limitations and short- application. Due to the non-linearity and non-stationarity of
comings, and the existing prediction methods are difficult to aircraft failure rate series, some scholars apply the idea of
achieve the ideal prediction effect. It can overcome the signal decomposition method in the field of signal pro-
shortcomings of traditional single prediction methods and cessing to aircraft failure rate prediction and process the
improve the accuracy of prediction. Therefore, some time series before prediction to reduce its non-smoothness.
scholars propose a combined model to predict the failure Then, optimize the model and propose a combined model
rate. based on decomposition integration. The combined model
The essence of combined model is to regard various of decomposition and integration is to decompose the
individual models as fragments representing different in- original data to generate different characteristic compo-
formation, disperse the unique uncertainty of individual nents. After these different components are generated, the
prediction, and reduce the overall uncertainty through the same or different models are used to predict. Finally, the
integration of information, so as to improve the prediction predicted values of each component are superimposed and
accuracy. Combined models can be divided into two types: integrated to form the final predicted values. For example,
multi-model combination and decomposition integration Wang et al. [42] proposed a fault prediction method based
combination. Multi-model combination combines two or on empirical mode decomposition (EMD) and least squares
more single models to complete the prediction of failure rate. support vector machine (LS-SVM). The effectiveness and
Yang et al. [33] proposed an aircraft failure rate prediction superiority of the proposed method are verified by using the
method based on the Holt–Winters seasonal model, ana- classical fault rate prediction example of Boeing aircraft.
lyzed the seasonal time series data, and analyzed the ap- Wang and Lu [43] proposed a failure rate prediction method
plication of the Holt–Winters seasonal model in aircraft based on EMD and RVM-GM model and selected the failure
failure rate prediction through an example. Wang and Yuan rate data of Boeing 757-700 aircraft as the research object for
[34] proposed an autoregressive model based on neural multiple sets of data within two years. The MAPE value
network residual correction, and the combined model was reached 8%, which effectively improved the prediction ac-
used to predict the failure rate of Boeing aircraft. Practice has curacy. Xu et al. [44] proposed a prediction method based on
proved that the model is suitable for short-term failure rate correlation vector empirical mode decomposition (RVEMD)
prediction. Celikmih et al. [35] collected maintenance and and grouped method of data handling (GMDH) recon-
failure data of aircraft equipment over a two-year period and struction. 700 kinds of aircraft have been researched on
employed a multi-layer perceptron (MLP) as a machine multiple sets of failure rate data within two years, and the
learning combination of artificial neural network (ANN), MAPE value reached 4.78%. The method based on RVEMD
support vector regression (SVR), and linear regression (LR). and GMDH reconstruction can well reflect the change law of
The algorithm was evaluated, and the results showed that the failure rate and has high prediction accuracy. The decom-
model improved the prediction accuracy of faults. Gu et al. position method has the first mock exam for the submodel
[36] proposed the interpolation fitting transfer learning after the breakdown rate of aircraft, but most of the existing
(ITF) algorithm of adaptive weight for the prediction of decomposition and integration models adopt unified model
4 Scientific Programming

prediction instead of selecting suitable models for each sub- constructing the final combined model, improving
mode. The complementary set empirical mode (CEEMD) the model’s explanatory ability and forecasting
method can effectively improve the smoothness of time series performance to meet the actual needs.
and has high adaptability. It can effectively solve the mode
The rest of this paper is organized as follows. Section 2
aliasing problem of the EMD method and the problem that the
introduces the theoretical introduction based on CEEMD
white noise added in the set empirical mode decomposition
and combined models, analyzes the modeling process in
(EEMD) cannot be neutralized. It can effectively solve the
detail, and gives the methods and steps of CEEMD, ARIMA
uncertainty of the decomposition level of the variational mode
model, and grey Verhulst model. Section 3 uses the ex-
decomposition (VMD) method, which shows the unique
perimental example to study the model. Through the
advantages of the CEEMD method. The CEEMD method can
collection of experimental data, the established model is
be used to decompose the aircraft failure rate prediction se-
used to carry out detailed research and discussion, and the
quence, and there is little research in this field.
evaluation index of prediction performance is given at the
The above researchers have carried out research on the
same time. Section 4 compares and analyzes the prediction
prediction of aircraft failure rate by the combined model.
results and accuracy of the established model with other
The application of combined model in various sciences
models in detail. Section 5 is devoted to discussion. Finally,
shows that the combined model can make full use of the
Section 6 gives the conclusions and corresponding
advantages of various methods and models and achieve good
suggestions.
prediction results. Compared with the single model, the
prediction results are improved to different degrees, and
better prediction performance is obtained. However, the 2. CEEMD and Combination Models
combined forecasting model also has certain shortcomings:
first, there is a certain blindness in the selection of fore- The aircraft mainly includes many parts such as the power
casting models. Since different forecasting models have their system, flight control system, hydraulic system, fuel system,
own applicable objects, choosing different models will have a and communication system. Among them, the fuel system,
certain impact on the subsequent forecasting results. Sec- as an important functional subsystem of the aircraft, un-
ondly, the weight coefficients of the combined forecasting dertakes the functions of stable and reliable fuel supply to the
model all use the same value, that is, the direct superposition engine, control of the position of the center of mass, and
method is used for the forecast values of different forecasting cooling of other systems. The aircraft fuel system is com-
models, and the key role of different weights on the com- posed of a fuel tank, actuator, fuel pipeline, and fuel control
bined model is not analyzed. Finally, the prediction accuracy system. If the fuel control system fails, the aircraft engine will
still needs to be further improved to achieve the ideal fail and the aircraft will fail, thus affecting the flight per-
prediction effect. Because the aircraft failure rate is affected formance and even the occurrence of safety accidents. In this
by various factors such as flight state, climatic conditions, paper, the failure rate of the aircraft fuel control system is
and comprehensive support capabilities and the cause of the predicted, aiming at predicting the failure of the fuel control
failure is highly uncertain, the selection of the modeling system through the predictive model, predicting the aircraft
method for the combined prediction model of the aircraft that may fail in time, conducting inspection, maintenance,
failure rate is still a current research focus. and health management in advance, and improving the
In view of this, in order to solve the existing deficiencies safety of the aircraft sturdiness and reliability.
and improve the prediction performance, this paper will
focus on the research on the failure rate prediction method
2.1. Modeling Process. In order to study the prediction model
of the combined model and propose a combined model to
of aircraft failure rate, the complementary set empirical
improve the failure prediction accuracy of the aircraft. The
mode decomposition (CEEMD) and combined model
major contributions of this research are as follows:
method are used, including ARIMA model and grey Ver-
(1) The CEEMD method is introduced to decompose the hulst model, to solve the coefficients of combined models by
time series of the collected aircraft failure rate. Using the entropy weight method, and finally a combined pre-
the decomposition and integration method, higher diction model is formed. The structure of the prediction
prediction accuracy, better directional prediction, model method is shown in Figure 1, which includes the
and higher robustness can be achieved, and the following three parts:
Hurst index is used to determine the appropriate
correlation prediction model for each decomposition (1) Select typical aircraft systems and fault collection:
component. select typical fuel systems from aircraft systems to
carry out research, specifically study the failure rate
(2) Select the ARIMA model and the grey Verhulst of the control system in the fuel system to predict,
model as the single-item model of the combined analyze the failure situation of an aircraft fuel control
forecasting model to perform combined forecasting system in detail, and collect the failure rate data; the
of the group failure rate. data are obtained from two aspects: on the one hand,
(3) Use an entropy weight method to solve the coeffi- the failure rate is collected from the historical fault
cients of the combined model to optimize the weight data over the years; then, the failure rate of the
coefficients of each forecasting model, thereby aircraft fuel control system is calculated, and the time
Scientific Programming 5

Selection of typical aircraft systems and collection of failure rates

Select aircraft typical system as research object

Historical failure data collection Fault information data collection and monitoring Computer data collection

Calculate the failure rate of typical aircraft systems

CEEMD and Combination Algorithm Implementation

CEEMD

IMF1 IMF2 ——— IMFm RF component Hurst index

calculation

ARIMA forecast ARIMA forecast ——— ARIMA forecast Grey forecast

Calculate the weight of each separation through the entropy weight method and obtain the corresponding
combined predicted value

Implementation and Verification of Combination Models

Accuracy Comparison and Analysis

Accuracy calculation verification and comparative SSE, MAE, RMSE, MSPE
analysis of each model , NRMSE, IA, MAPE

Uncertainty and stability analysis

Figure 1: Flowchart of failure rate prediction.

series of failure rate is obtained, and the failure entropy weight method, the separated coefficients
prediction research is carried out with the percentage are solved to optimize the weight coefficients of each
of failure rate as the overall failure rate. prediction model, and the combined prediction
(2) Implementation of complementary set empirical model is deduced to obtain the final prediction re-
mode decomposition and combination algorithm. sults, which provides an effective model for pre-
Firstly, IMF fluctuation term and RF trend term of dicting the aircraft failure rate.
fault rate series are separated by CEEMD, and the (3) Accuracy analysis and comparison of the model: in
single prediction model is determined by Hurst order to evaluate the prediction performance of the
index calculation. Secondly, the ARIMA prediction combined model, it is compared with other models.
model of each IMF component and the grey Verhulst Using a variety of accuracy indicators as evaluation
prediction model of RF trend term component are criteria, compare the accuracy of other model vali-
established. Finally, the combined model based on dation data with the results of the proposed com-
each ARIMA and grey Verhulst model is realized to bined prediction model, carry out systematic
predict the failure rate value (as the test data pa- comparative analysis and research, and carry out
rameter) and failure rate value (as the output pa- specific application and verification to achieve the
rameter). Based on the combined modeling theory of corresponding prediction effect.
6 Scientific Programming

2.2. CEEMD. CEEMD is an improved algorithm based on Therefore, the original signal sequence is finally
EMD and EEMD; the empirical mode decomposition decomposed into S(n) � 􏽐K
k�1 IMFk + R(n).
(EMD) algorithm proposed by Huang et al. in 1998 is a
method for dealing with non-linear signals, which has
shortcomings such as modal aliasing and end-point effects. 2.3. Hurst Index. There is a known time series x1 , x2 , . . . , xn .
In view of the shortcomings of EMD, Huang et al. proposed The average value is xn � 1/n 􏽐ni�1 xi , and the standard de-
􏽱��������������
ensemble empirical mode decomposition (EEMD) in 2009, viation is Sn � 1/n 􏽐ni�1 (xi − xn ). For 1 ≤ t ≤ n, the cumu-
which added white noise sequences on the basis of EMD to
lative deviation is xt,n � 􏽐ti�1 (xi − xn )2 , and range is
overcome the modal aliasing that appeared in EMD.
However, in each decomposition process, adding dissimilar R � max􏼈xt 􏼉 − min 􏼈xt 􏼉. The Hurst index refers to the co-
1≤t≤n 1≤t≤n
white noise makes EEMD have the problems of noise res- efficient H satisfying the following equation [46], where C is
idue, many iterations, and slow operation efficiency. In 2010, a constant. By specifying a value of C (e.g. 0.5), the Hurst
Yeh et al. [45] proposed CEEMD based on EEMD. Different index can be obtained.
from EMD and EEMD, CEEMD uses adding different white
noise to the signal and EMD processing, respectively, to
remove the adverse effect of adding white noise to EEMD, 2.4. ARIMA Model. The autoregressive integrated moving
reduce reconstruction error as much as possible, overcome average (ARIMA) model was proposed by Box and Jenkins
the noise residue, and improve the calculation efficiency, and [47]. ARIMA consists of three parts, namely, AR, I, and MA,
the effect will be better when it is used for signal decom- in which AR is the autoregressive model, I represents the
position. S(n) is defined as the original sequence, and Ek (·) difference, and MA is the moving average model. As a kind
and IMFk represent the K-th modal components generated of time series analysis method, the ARIMA model can
by EMD and CEEMD, respectively. The specific steps are as predict future values according to historical values. ARIMA
follows: is the most widely and commonly used time series pre-
diction. Its modeling is easier than more complex methods.
Step 1: CEEMD makes I test on the signal, and the first Because of its simplicity and stability, it has been widely used
modal component is obtained by EMD algorithm and in forecasting, social, economic, and engineering fields. The
recorded as general expression of the ARIMA model is
1 I
IMFi (n) � 􏽘 IMFi1 (n). (1) Φ(L)(1 − L)d yt � ε + Θ(L)εt . (6)
I i�1
Among them, Φ(L) � 1 − λ1 L − λ2 L − . . . − λP LP is the
Step 2: when k � 1, the current unique margin is ob- P-order autoregressive coefficient polynomial, Θ(L) � 1 +
tained and recorded as θ1 L + θ2 L + . . . + λq Lq is the P-order moving average coef-
ficient polynomial, L is the lag operator, λ and θ are the
r1 (n) � S(n) − IMF1 (n). (2) estimated values of the respective variables, yt � c + yt−1 +
μt is a d-order single integer sequence, c is a constant, μt is a
Step 3: conduct I tests (i � 0, 1, 2, . . . , I) and decom- stationary sequence, t � 1, 2, . . . , T, and εt is a white noise
pose continuously r1 (n) + ε1 E1 (]i (n)) until the first sequence with a mean of 0 and a variance of σ 2 . The ARIMA
modal component is obtained. On this basis, calculate model prediction steps are as follows:
the second modal component:
Step 1: stationarity test.
1 I
IMF2 (n) � 􏽘 E1 􏼐r1 (n) + ε1 E1 􏼐]i (n)􏼑. (3) The stability of the data is preliminarily judged
I i�1 according to the sequence diagram, autocorrelation
function diagram, and partial autocorrelation function
Step 4: for each of the remaining stages, that is, diagram of the time series.
k � 2, 3, · · · , K, according to Step 3 above, calculate the
Step 2: smoothing.
k-th residual signal and the (k + 1)-th modal component:
Stationarize the non-stationary time series data until
rk (n) � rk−1 (n) − IMFk (n),
the processed data can pass the stationarity test.
I (4)
IMFk+1 (n) � 􏽘 E1 􏼐rk (n) + εk Ek 􏼐]i (n)􏼑􏼑. Step 3: model identification and order determination.
i�1 Establish a corresponding time series model according
to the identified features. After smoothing, if the partial
Step 5: repeat Step 4 until the residual signal can no autocorrelation function is truncated and the auto-
longer be decomposed, then the algorithm stops, and correlation function is tailing, the AR model is estab-
the final residual signal is lished; if the autocorrelation function is tailing, the
autocorrelation function is tailing; if the function is
truncated, the MA model is established. If both the
K
partial autocorrelation function and the autocorrela-
R(n) � S(n) − 􏽘 IMFk . (5) tion function are tailed, the sequence is suitable for the
k�1
ARMA model.
Scientific Programming 7

Step 4: model parameter estimation. −1

a􏽢 � 􏼐BT B􏼑 BT Y. (10)
Estimate the model parameters and test the estimated
values to determine the reliability of the parameter Then, the solution of the whitening equation (time
estimates. response function) is
Step 5: model checking.
ax(1)
In the white noise test, it is assumed that the model x(1) (t) � 0
. (11)
judges whether the residual sequence is a white noise bx(1) (1) at
0 + 􏼐a − bx0 􏼑e
sequence.
Step 6: prediction of the model. The time response sequence for the grey Verhulst
model is
After building the ARIMA(p, d, q) model, a predictive
model is formed to predict future values based on historical ax(1)
􏽢 (1)
x k+1 �
0
. (12)
values. bx(1) (1) ak
0 + 􏼐a − bx0 􏼑e

x(1) (0)
0 will be taken as x1 ; then, (11) becomes
2.5. Grey Verhulst Model. The grey system theory was first
proposed by Professor Deng Julong to deal with the “small ax(0)
sample, poor information” system. The grey prediction 􏽢 (1)
x k+1 �
1
. (13)
bx(0) (0) ak
1 + 􏼐a − bx1 􏼑e
method is based on the grey system theory and is a method
for fuzzy prediction of the system by establishing approx-
Step 4: accumulate and restore the prediction model.
imate differential equations. It includes GM (1, 1), GM (1, n),
GM (0, n), Verhulst, and other models. The grey Verhulst x􏽢 (1)
k+1 carries out the cumulative reduction and reduction
model is an important part of the grey system theory and is a operation, and the prediction model of the grey Verhulst can
single sequence first-order non-linear dynamic model [48]. be obtained as
The grey Verhulst model prediction steps are as follows:
􏽢 (0)
x 􏽢 (1)
k+1 � x 􏽢 (1)
k+1 − x k .
(14)
Step 1: carry out the accumulation and generation
operation.
Let the original sequence be known as x(0) � (x(0) 1 ,
x(0)
2 , . . . , x (0)
n ) and perform a cumulative generation 2.6. Solving the Coefficients of the Combined Model by the
(1-AGO) on the original sequence to obtain a new Entropy Weight Method. Based on CEEMD, the failure rate
sequence x(1) � 􏽮x(1) (1) (1)
1 , x2 , . . . , xn 􏽯, where
sequence is decomposed into IMF term and RF residual term
k
of each order and then predicted according to the charac-
x(1)
(0)
(k � 1, 2, . . . , n). teristics of each component sequence, and the predicted
k � 􏽘 xi , (7)
i�1 values of the components are added to obtain the final
prediction result. However, this method is likely to cause
Step 2: generate a sequence of adjacent mean values. omission or loss of some information, thereby affecting the
x(1) is generated as the next-to-neighbor mean Z(1) accuracy of prediction. Therefore, it is necessary to take into
� (z(1) (1) (1) (1) (1) (1) account the physical meaning behind each component and
1 , z2 , . . . , zn ), where zk � 1/2(xk + xk−1 ),
k � 2, 3, . . . n. systematically mine the implicit information hidden in the
IMF component and the RF remainder to determine the
Then, the grey Verhulst model is [49] importance of each factor. At the same time, different single-
2 item models also have different predictive capabilities, so the
x(0) + az(1) � b􏼐z(1) 􏼑 . (8)
weight coefficient of the single-item models should not
simply take the average of the number of single-item models
The corresponding whitening equation is
in combined forecasting but give different weights according
dx(1) 2 to the predictive ability of different single-item models;
+ ax(1) � b􏼐x(1) 􏼑 . (9)
single-item models with strong predictive performance are
dt
assigned larger weights. Considering the above two reasons,
In formulas (7) and (8), a and b are parameters and t is an intelligent combination model based on entropy weight
time. method index weight calculation is constructed, so that the
Step 3: establish an approximate time response sequence. prediction of the combination model is more accurate, the
2 generalization ability of the model is strengthened, and the
−z(1)
2 (z(1)
2 ) ⎤ x(0)
2 ⎤
⎢
⎡
⎢
⎢ (1) (1) 2 ⎥⎥⎥ ⎢
⎡
⎢
⎢ (0) ⎥
⎥⎥⎥ purpose of improving the prediction accuracy is finally
⎢
⎢ −z (z ) ⎥
⎥
⎥ ⎢
⎢ x 3 ⎥ ⎥⎥⎥. Then, the
Make B � ⎢ ⎢
⎢ 3 3 ⎥
⎥
⎥ and Y � ⎢
⎢
⎢ achieved.
⎢
⎢
⎣ ⋮ ⋮ ⎥⎦ ⎣ ⋮ ⎥⎥⎦
⎢
⎢
The basic idea of the entropy weight method is to use the
2
−z(1)
n (z(1)
n ) x(0)
n discrete degree of the index, that is, the information entropy,
least squares estimation parameters of the grey Ver- to measure the importance of the index [50]. The weight of
hulst model are listed as each index is determined by applying the entropy weight
8 Scientific Programming

ARIMA model training

Prediction result of IMF

ARIMA
component failure rate

The coefficients solved by

Aircraft failure rate data CEEMD the entropy weight method
Failure rate
Verhulst model training are used to establish a
combined model

Grey Verhulst RF residual failure rate prediction results

Figure 2: Execution steps of the combined model.

method. Based on the objective data, the index weight is method is used to solve the coefficient, the basis of the
determined according to the entropy value of the data. The combination weight is determined, and the final combination
weight of the indicators can represent the differential impact function is obtained, and the corresponding accuracy eval-
of objective data on the results. By using the entropy weight uation index is used for comparison and analysis, and the
method, calculate the weights of the predicted values cor- overall prediction of the failure rate is completed.
responding to IMF1, IMF2, IMFm, and RF decomposed by
CEEMD. The entropy weight method determines the weight
steps as follows:
3. Experimental Example
Step 1: data normalization. 3.1. Experimental Example Dataset
To quantify the high-frequency and low-frequency data 3.1.1. Composition and Failure Factor Analysis of Aircraft
decomposed by CEEMD, assuming that there are m Fuel Control System. As an important part of the aircraft
evaluation indicators and n evaluation objects, the system, the aircraft fuel control system plays a key role. Its
original data matrix is A  (aij )m×n; after normaliza- composition and fault factors are shown in Figure 3. The
tion, the matrix is obtained on B  (bij )m×n, aircraft fuel control system generally includes two parts:
bij  aij /maxaij . fuel supply control system and centroid position control
Step 2: calculation of entropy. system. The fuel supply control system controls the specific
The entropy of the i-th index is hi  −k nj1 fij ln fij , fuel consumption of each fuel tank according to the fuel
where fij is the standardized value of the index pro- consumption of the engine, so as to meet the fuel con-
portion of the j-th evaluation object under the i-th sumption demand of the engine; the centroid control
index, fij  bij / nj1 bij , and k is the adjustment co- system controls the centroid position of the aircraft
efficient, k  1/ln n. according to the needs, reduces the control burden of the
control surface, and ensures that the centroid position of
Step 3: calculation of entropy weight.
the aircraft is basically unchanged. The fuel supply control
After calculating the entropy of the i-th indicator, the system and centroid control system can calculate the air-
entropy weight of the i-th indicator is calculated as craft mass and centroid position information in real time
ωi  1 − hi /m − m i1 hi , (0 ≤ ωi ≤ 1, i1 ωi  1),
m
and through the integration of mass and centroid position
T
ω  [ω1 , ω2 , . . . ωm ] is the weight of the corresponding calculation module, which can be provided to the aircraft
prediction values of IMF1 and IMF2. IMFm and RF were dynamic system and other subsystems to meet their needs.
decomposed by CEEMD. The ARIMA model was set from The fuel supply control system is composed of a converter,
IMF1 to IMFm. The predicted value at time t is fuel control distributor, and fuel tank control unit. The

X(t)  m i1 (ωi xi (t)), i1 ωi  1.
m
centroid position control system includes a converter, fuel
The execution steps of the combined model are shown in control switch, and fuel tank control unit. According to
Figure 2. First, the data of the aircraft failure rate are collected statistical analysis, there are many factors causing the
in detail, with the failure rate percentage data as the bench- failure of aircraft fuel control system. The factors affecting
mark, and the collected dataset is divided into two parts: the failure rate include the following aspects.(1) Flight
training data and test data. Secondly, using complementary mission: including the length of flight, the number of take-
ensemble empirical mode decomposition (CEEMD) aircraft off and landing, and the number of missions. If the flight
failure rate data, the ARIMA model and grey Verhulst model mission is heavy, the probability of failure will be greater.
are used for training, and the failure rate is predicted to obtain (2) External environmental conditions: mainly including
the corresponding actual value and predicted value of failure. weather conditions, ambient temperature, humidity, and
Again based on the ARIMA and grey Verhulst forecasting other factors. Abnormal weather conditions and temper-
models, the predicted failure rate values are treated as input ature and humidity will have a certain impact on the failure
variables to the combined model, which forms the model to probability of fuel control system. The worse the weather
generate input and output datasets. Finally, the entropy weight conditions meaning that the environment beyond the
Scientific Programming 9

Maintenance
quality

Converter

Fuel supply control system Fuel control distributor

Aircraft fuel Each fuel tank control unit

External product
control
environment Converter quality
system
Center of Mass Position Fuel control switch
control system
Fuel tank control unit

Flight mission

Figure 3: Composition and fault factors of fuel control system.

reasonable temperature and humidity range, the greater the 6.0

probability of failure. (3) Product quality: devices with
good product quality, high system integration, and high 5.5
reliability will reduce the failure rate of fuel control system 5.0
to a certain extent. At the same time, products with poor
4.5
Failure rate (%)

quality and low stability will also increase the failure rate of
fuel control system.(4) Maintenance quality: when the fuel 4.0
control system is in use, it needs regular maintenance and
3.5
corresponding maintenance to improve the system reli-
ability and reduce the failure rate. In some previous aircraft 3.0
failure rate prediction studies, many factors such as flight
2.5
time, ambient temperature, ambient humidity, product
quality, and support ability need to be comprehensively 2.0
considered, which increases the complexity of the pre- 2014/03 2014/10 20150/6 2016/02 2016/10 2017/06 2018/02
diction model. At the same time, there are many coupling Date
relationships between these influencing factors, and some
Figure 4: Failure rate timing diagram.
influencing factors are difficult to obtain accurately in
practice. Based on this, this paper will only consider the
main failure rate of aircraft fuel control system to be compared with the real data. The collected failure rate time
predicted, without considering other influencing factors. sequence is shown in Figure 4, and the statistical parameters
of the dataset including average value, standard deviation,
maximum value, minimum value, skewness, kurtosis, and
3.1.2. Fault Data Acquisition and Analysis. The failure data median value are obtained at the same time. As shown in
of aircraft fuel control system are tested and evaluated by Table 2, combined with Figure 3 and Table 2, it can be seen
collecting the failure data of aircraft fuel control system for that the failure rate data of the fuel control system fluctuate
48 months from March 2014 to February 2018. The data are in a smaller range, which is a small sample, non-linear, and
divided into two subsets: the first 80% of the data samples are non-stationary curve. The maximum failure rate is 5.7%, the
used as the training dataset, and the last 20% are used for the maximum failure rate data does not exceed 6%, the mini-
test dataset. The training dataset is used for model training, mum value is 2.1%, and the minimum failure rate is not less
and the test dataset is used to evaluate the performance of the than 2%. The whole failure rate basically fluctuates between
established prediction model to verify the effectiveness of the 2% and 6%, and the average failure rate is 3.7%. For the
proposed model. Using the data of the first 38 months of the aircraft fuel control system, the failure rate of nearly 6% is
sample from March 2014 to April 2017, the prediction model relatively high which directly affects the operating state of
is established as the modeling input, the prediction is verified the engine and has a certain impact on the flight safety of the
by using the failure rate of the last 10 months from May 2017 aircraft. Therefore, it is necessary to effectively predict the
to February 2018, and then the prediction results are failure rate of aircraft fuel control system, improve the
10 Scientific Programming

Table 2: Statistical parameters of failure rate dataset.

Dataset Maximum Minimum Variance Median Standard deviation Range
5.7 2.1 1.097 3.85 1.047 3.6
Kurtosis
48-month failure rate data Average value Mean standard error Deviation Standard deviation Kurtosis
standard error
3.72 0.151 −0.016 0.343 −1.143 0.674

operation state of aircraft engine, and ensure the safety and Table 3: Predictive model capability formula.
reliability of the aircraft system.
Evaluation
Formula
parameters
3.2. Prediction Performance Evaluation Indicators. To eval- SSE 􏽐ni�1 (yi − y􏽢 )2
n
uate the prediction performance of the prediction model, we MAE 1/n 􏽐i�1 |􏽢
yi − yi |
􏽱���������������
need to measure and analyze the prediction results, and the n
RMSE 1/n 􏽐i�1 (􏽢yi − yi )2
emphasis and adaptability of different evaluation indicators 􏽱��������������
are also different. At present, a variety of evaluation indi- MSPE 1/n 􏽐ni�1 (􏽢yi − yi /yi )2
􏽱���������������
cators are proposed [51]. In order to evaluate the model and NRMSE 1/n 􏽐ni�1 (􏽢
yi − yi )2 /y
test the effectiveness of the method from many aspects, this
IA 1 − 􏽐ni�1 (􏽢
yi − yi )2 / 􏽐ni�1 (|􏽢
yi − y| + |yi − y|)2
paper selects the following seven evaluation criteria to
measure the error between the predicted value and the real MAPE 1/n 􏽐ni�1 |􏽢
yi − yi /yi | × 100%
value, specifically sum of squares (SSE), mean absolute error
(MAE), root mean square error (RMSE), mean square
established as the modeling input, and the CEEMD model
percentage error (MSPE), normalized root mean square error
was used to decompose the data. After the decomposition of
(NRMSE), consistency index (IA), and mean absolute per-
the obtained CEEMD, the timing is shown in Figure 5.
centage error (MAPE) are used as reference. Using the above
It can be seen from Figure 4 that the failure rate data of
evaluation indicators to comprehensively measure and
the first 38 groups of data are decomposed into 5 compo-
evaluate the prediction effect, it has a systematic and com-
nents, including 4 IMF fluctuation terms and 1 RF trend
prehensive prediction ability compared with general indi-
term. Among them, the frequency and amplitude of IMF1
cators. Suppose the predicted value is y 􏽢 � 􏼈y􏽢1, y
􏽢 2, . . . , y
􏽢 n 􏼉,
and IMF2 components are still at a high level, but the
the true value is y � 􏼈y1 , y2 , . . . , yn 􏼉, y is the average value of
fluctuation is significantly lower than the original data, and
the true value, and the corresponding prediction model
the smoothness is significantly improved. The frequency and
capability formula is obtained as shown in Table 3.
amplitude of the trend items IMF3 and IMF4 gradually
The SSE calculates the sum of squares of the errors
decreased, and the sequence gradually became stable. The
between the fitted data and the corresponding points of the
trend term RF shows a plateau in the failure rate over the first
original data. The smaller its value, the better the selection
38 months. The failure rate data decomposed by CEEMD are
method and fit of the model and the better the data pre-
more regular, and the change trend of the overall failure rate
diction effect. When the predicted value is in good agree-
range after decomposition is consistent with the change
ment with the actual value, the MAE is the expected value of
trend of the actual aircraft system failure rate. Each com-
the absolute error loss, and its smaller value means higher
ponent still contains the characteristic information of the
prediction accuracy. RMSE is also one of the comprehensive
original failure rate sequence, which reduces the interference
indicators of error analysis, representing the degree of
of abnormal data in the modeling process, smoothes the
dispersion of the predicted value, and the best fit is
curve, facilitates data prediction, and can effectively improve
RMSE � 0. MSPE reflects the dispersion degree of error. The
the accuracy of data prediction. The obtained fluctuation
smaller the value, the smaller the error fluctuation to a
terms and trend terms are also more representative for
certain extent. The smaller the NRMSE, the higher the ac-
reflecting the real physical components of the original data,
curacy. IA represents the change trend and degree of
and based on this, the next step of predictive modeling
consistency between the predicted value and the measured
analysis is carried out. Use Section 2.3 to calculate the Hurst
value, and the closer it is to 1, the higher the trend and
index of the five components decomposed by CEEMD to
consistency are. The MAPE can be treated as a benchmark,
obtain the Hurst index of each component, as shown in
and it is generally considered that when the MAPE is less
Table 4.
than 10%, the prediction accuracy is higher.
The self-similarity of time series can be measured by the
Hurst index. When 0 ≤ H < 0.5 means negatively correlated,
3.3. Prediction Model Implementation the time series has inverse persistence. When H � 0.5 means
that the time series is purely random, the current state does
3.3.1. CEEMD. Using the collected fuel control system not affect the future state, and the time series is irrelevant.
failure rate data for the 38 months before the period from When 0.5 < H ≤ 1 represents a positive correlation, it means
March 2014 to April 2017, a prediction model was that the time series has persistent behavior and long
Scientific Programming 11

2
IMF1

-2
2014/03 2015/09 2017/04
Time
1
IMF2

-1
2014/03 2015/09 2017/04
Time
0.5
IMF3

-0.5
2014/03 2015/09 2017/04
Time
0.5
IMF4

-0.5
2014/03 2015/09 2017/04
Time
5
RF

3
2014/03 2015/09 2017/04
Time
Figure 5: Timing diagram after CEEMD.

memory, so the larger the Hurst index, the higher the self- Table 4: Hurst index of each component.
similarity of the time series. It can be seen from Table 4 that
Components Hurst index
the Hurst exponent values of IMF1, IMF2, IMF3, and IMF4
are between 0.5 and 0.9, indicating that the corresponding IMF1 0.5682
components are periodic nonstationary. The ARIMA model IMF2 0.5694
IMF3 0.8842
can effectively analyze the correlation of periodic non-sta-
IMF4 0.8853
tionary data series and is suitable for time series. It is rea- RF 0.9123
sonable to choose the ARIMA model as the prediction model
of IMF1 to IMF4 stationary characteristics. However RF has
a large Hurst exponent, and its exponent value is greater
than 0.9, indicating that the RF component has stable 3.3.2. ARIMA Model. According to the basic principle in
changes and certain non-linear characteristics. Selecting the Section 2.4, after decomposing the collected 38 groups of
grey Verhulst model has a good predictive ability for non- failure rate data by CEEMD, the obtained IMF1 to IMF4 data
linear sequences and is used as a prediction model for RF sequences are converted into stationary sequences through
components. Also, from the perspective of fluctuation, the several differential operations, and the order of each model
aircraft failure rate will increase significantly at the begin- can be determined. With the help of SPSS software, it can be
ning and will gradually decline in the future. When the obtained that IMF1 to IMF4 are transformed into stationary
failure rate reaches saturation, the growth rate is close to series after one difference operation, at the same time, the
zero, and there is a negative growth with time. The failure appropriate parameter values are calculated by SPSS soft-
rate shows an “s” downward trend with the development of ware to screen P and Q, and the optimal parameter values are
time and the sample size of failure rate is small, which is selected according to AIC, so as to determine the optimal
more suitable for the grey Verhulst model and has certain model. Finally, it is determined that the optimal models of
advantages over the ARIMA model. The grey model has failure rate prediction model are IMF1-ARIMA (3, 1, 1),
good prediction effect, so the RF term decomposed by IMF2-ARIMA (2, 1, 3), IMF3-ARIMA (2,1,4), and IM4-
CEEMD is used for prediction. Therefore, the modeling ARIMA (2, 1, 2), and so far the order and parameter se-
method is more reasonable. lection of each prediction model is completed, the prediction
12 Scientific Programming

Combination Model Failure Rate Prediction

Weight factor 1 Weight factor 2 Weight factor 3 Weight factor 4 Weight factor 5

Using the entropy weight method to solve each weight coefficient

Predicted value 1 Predicted value 2 Predicted value 3 Predicted value 4 Predicted value 5

ARIMA(IMF 1) ARIMA(IMF 2) ARIMA(IMF 3) ARIMA(IMF 4) Verhulst(RF)

Figure 6: The process of solving the weight coefficient of the entropy weight method.

model is initially established, and the model can be tested in complexity, based on the entropy weight method combined
the next step. Establish a dependent prediction model. Based modeling theory, the corresponding weight coefficients are
on the analysis of IMF1 to IMF4 models based on the above obtained, as shown in Table 5.
models, get the corresponding mathematical model, predict All IMF variables decomposed by CEEMD participate in
the subsequent 10 groups of data, and get the predicted value the construction of the model, and the weight coefficients of
of IMF1 to IMF4 fluctuation term. the second, third, and fourth IMF components are only
small, indicating that the contribution rate of the IMF
component to the final result is small, which may represent
3.3.3. Grey Verhulst Model. According to the basic principle
the fluctuation of failure rate caused by some random
of Section 2.5, the collected RF data of 38 groups of failure
factors. The weight coefficients of the first and fifth IMF
rates after CEEMD are decomposed, and the corresponding
components are relatively large, indicating that this com-
prediction model data are obtained by using the grey system
ponent may represent some important factors affecting the
modeling software of China Southern Airlines (V7.0) to
failure rate. The final prediction model is obtained by
calculate a  −0.06, b  −0.01, the corresponding grey Ver-
multiplying the corresponding predicted value by the cor-
hulst model is dx(1) /dt − 0.06x(1)  −0.01(x(1) )2 , and the
 (1)
responding weight coefficient, and the combined model
 t  ω1 y
 IMF1 + ω2 y
 IMF2 + ω3 y
 IMF3 + ω4 y
 IMF4 +
−0.06k
corresponding time is x k+1  −0.19/ − 0.04 − 0.02e . A
expression is y
 RF . Through the combined model, the predicted values of
dependent prediction model is established, the RF com-
ω5 y
ponent is modeled based on the above model, the subse-
the last 10 groups of test data are obtained. Compared with
quent 10 sets of data are predicted, and the predicted value of
the simple addition method, the entropy weight method has
the RF component is obtained.
good interpretability and better prediction accuracy.

3.3.4. Entropy Weight Method. The entropy weight method 4. Comparison and Analysis of Combined
weight coefficient solution and construction process is Model Results
shown in Figure 6.
Most of the previous models use simple addition to Using ARIMA model, CEEMD-ARIAM model, CEEMD-
aggregate the predictions of all models into a final predic- grey Verhulst model, and the proposed CEEMD and
tion, while the proposed model utilizes the grey Verhulst combined prediction model to predict 10 sets of failure rate
model for RF components simultaneously after utilizing the data from May 2017 to February 2018, the change trends of
ARIMA model for each prediction model of the IMF1-IMF4 actual failure rate and corresponding failure rate prediction
components. After modeling and predicting, the corre- values of the four models are shown in Figure 7.
sponding predicted values are obtained, which are based on In Figure 7, it can be seen that the fitting trend of the 10
the predicted values of 5 groups of corresponding failure sets of failure rate data from 2017/05 to 2018/02 after using
rates. Assuming that the ARMIA model is used to predict the a single ARIMA model is not high compared with the
failure rate of IMF1 as y  IMF1 , the ARMIA model is used to actual failure, and the prediction effect of the failure rate by
predict the failure rate of IMF2 as y  IMF2 , the ARMIA model the ARIMA model is not good. Ideally, in the whole failure
is used to predict the failure rate of IMF3 as y  IMF3 , the rate prediction period, the deviation of the front section is
ARMIA model is used to predict the failure rate of IMF4 as large, and the deviation of the rear section is small, but the
 IMF4 , and the grey Verhulst model predicts the failure rate
y overall fitting degree is not high. The fitting degree of the
obtained by RF as y  RF ; on the basis of not increasing the CEEMD-ARIMA model tends to be reasonable, which is
Scientific Programming 13

Table 5: Weight coefficients of each model. 5.0

Weight factor Value

4.5
ω1 0.211
ω2 0.198 4.0
ω3 0.038

Failure rate (%)

ω4 0.049 3.5
ω5 0.504
3.0

significantly different from that of single the ARIMA 2.5

model. In the front and middle stages, the fitting effect is
significantly better than that of the single ARIMA model, 2.0
and the fitting degree has been significantly improved, but
the overall effect is not ideal. The CEEMD-Verhulst model 2017/05 2017/09 2018/02
is used for prediction. The predicted value of the failure rate Date
and the actual value have the largest fitting deviation ARIMA model
among the four models. The fitting degree is not high in the CEEMD-ARIMA model
entire failure rate prediction period, and the overall fitting CEEMD-Verhulst model
effect compared with other three models is not ideal. That CEEMD-Combined model
CEEMD and combined model are used to predict the Actual failure rate
failure rate, and compared with other models, it has the Figure 7: Comparison between predicted value and actual value of
highest fitting degree, the deviation between the actual failure rate of different models.
value and the predicted value is the smallest, and the
prediction effect is good, which can meet the prediction
needs of aircraft failure rate. 8

In order to select the best prediction model, we com- 7

pared the error indexes of the validation data in different 6.58
6.01
models, selected 10 groups of validation samples for error 6
calculation, and obtained the corresponding SSE, MAE,
RMSE, MSPE, NRMSE E, IA, and MAPE which are shown in 5

Figures 8–10.
value

4
As can be seen from Figure 8, the CEEMD combined
model performs best among all models, mainly reflecting 3
the following: the SSE value of the model is 0.84, MAE is
0.22, and RMSE is 0.28, which is the smallest compared 2 1.67
with ARIMA model, CEEMD-ARIMA model, and 1 0.84 0.77 0.81
0.65 0.67
CEEMD grey Verhulst model. The SSE, MAE, and RMSE 0.37 0.22
0.41 0.28
values of the CEEMD-grey Verhulst model are the largest, 0
indicating that the prediction accuracy of the CEEMD- SSE MAE RMSE
grey Verhulst model is not high. Compared with the ARIMA Model CEEMD-Verhulst Model
CEEMD-grey Verhulst model, the SSE, MAE, and RMSE CEEMD-ARIMA Model CEEMD and Combined Model
indexes of this model are reduced by 87.2%, 67.1%, and Figure 8: Comparison of SSE, MAE, and RMSE of different
65.4%, and the model has higher prediction accuracy than models.
other models.
It can be seen from Figure 9 that the MSPE of this model
is 0.17 and the NRMSE is 0.12, which is lower than that of model. The prediction of ARIMA model and CEEMD grey
other models, and the MSPE index of this model is reduced Verhulst model is reasonable, while the prediction of
by 68.5% compared with the ARIMA model. Compared with CEEMD-ARIAM model is good, but the overall prediction
the index, the decrease range is larger, and the difference performance is not high, while the MAPE value of the
between the two is 0.2. At the same time, the IA of this model CEEMD combined model is 8.36%. Compared with other
is 0.82. The consistency index (IA) of this model is larger models, it is the smallest and less than 10%, which also shows
than that of other models, and compared with other models, that the prediction accuracy of the combined model is high.
it is closer to 1. Compared with the ARIMA model, the IA The results of the above seven evaluation indexes show the
index has increased by 26.3%. From the above indexes, the prediction effect of the combined model. Therefore, the
prediction performance of the model is better. CEEMD combined model proposed in this paper is suitable
It can be seen from Figure 10 that the MAPE values of for aircraft failure rate prediction and improves the per-
the ARIMA model, the CEEMD-ARIAM model, and the formance and accuracy of aircraft failure rate prediction. In
CEEMD-grey Verhulst model are all higher than 10%, which order to better evaluate whether the prediction accuracy of
are two to three times the value of the CEEMD-combination the proposed prediction model is significantly better than
14 Scientific Programming

1.0 different models and the accuracy indicators are shown in

Table 6.
0.82 Table 6 provides a comprehensive comparison in terms
0.8
of prediction accuracy metrics between the proposed model
and the EMD-ARIMA model, the BP model, and the triple
0.62
0.6 exponential smoothing model. Judging from the accuracy of
0.54 0.52 different evaluation indicators, the IA value of the proposed
value

model is higher than that of other models, while the other six
0.4 indicators are lower than those of other models. It shows that
0.31 0.32
0.28 compared with other models, the prediction model can
0.22 predict the aircraft failure rate more accurately and has
0.2 0.17 0.16
0.12
better prediction performance. It can be seen from the di-
rectional statistics that the combined model has the highest
0.03
0.0
value of Dstat among the four models, indicating that the
MSPE NRMSE IA model has better performance in the direction of failure rate
prediction. The EMD-ARIMA model and the BP model have
ARIMA Model CEEMD-Verhulst Model
CEEMD-ARIMA Model CEEMD and Combined Model higher values of Dstat than the cubic exponential smoothing
model. In addition, the combined model is also much better
Figure 9: Comparison of MSPE, NRMSE, and IA of different than other prediction models, and the evaluation result of
models.
Dstat has increased from 72.32%–84.78% to 99.52%. Using
the evaluation index of Dstat as the evaluation standard
45 related to maintenance decision-making has more practical
guidance and constructive significance for the direction
judgment of corresponding decision makers in aircraft fault
36
prediction and maintenance. From the perspective of sta-
tistics, the prediction accuracy of this model is significantly
27
27.31 26.85 better than that of other models. The DM test is used to
evaluate the proposed model, and the statistical data and P
value

values (in parentheses) of the target model and benchmark

18
15.46
model are obtained as shown in Table 7.
It can be seen from Table 7 that at the 5% significance
level, the proposed CEEMD combined model is significantly
9 8.36
better than the EMD-ARIMA model, the BP model, and the
triple exponential smoothing model. The DM test results
0
further confirm the superiority of the model from a sta-
MAPE (%) tistical point of view.
ARIMA Model CEEMD-Verhulst Model Because the aircraft failure rate has the characteristics of
CEEMD-ARIMA Model CEEMD and Combined Model uncertainty and instability, it is necessary to evaluate the
Figure 10: Comparison of MAPE of different models.
uncertainty and stability of the established model. The
uncertainty of the prediction model mainly includes input
uncertainty, parameter uncertainty, and model uncertainty,
other models, this study introduces a directional statistic which will have a certain impact on the model. In order to
Dstat to evaluate the directional prediction ability of the verify the uncertainty of the proposed model, the coefficient
model. of variation is used for verification. Coefficient of variation is
a statistical index to measure the dispersion degree and
1 n
Dstat   α × 100%, (15) variation degree of each observation value. If the ratio of
n i1 i standard deviation to average is taken as the coefficient of
variation, CoV  σ/μ, σ is the standard deviation, and μ is
yi − yi−1 ) ≥ 0;
where αi  
1, (yi − yi−1 ) × ( the sample mean value [56]. The coefficient of variation can
yi is the
0, otherwise, eliminate the influence of different units or averages on the
actual value, and y  i is the predicted value. The larger the comparison of two or more degrees of variation. It is a
value of Dstat , the higher the prediction accuracy and the dimensionless absolute value, indicating the dispersion of
better the direction prediction performance. Also, we use random variables around its mean. The coefficient of vari-
the Diebold–Mariano (DM) test to test the pros and cons ation of different models is obtained by calculation as shown
of the model from a statistical point of view [52]. At the in Figure 11.
same time, in order to further verify the accuracy of the As can be seen from Figure 11, the coefficient of variation
CEEMD-combined model, the EMD-ARIMA model [53], of the proposed CEEMD combined model is 0.23, which is
the BP model [54], and the triple exponential smoothing smaller than that of the BP model, EMD-ARIMA model,
model [55] were used for comparison and analysis. The cubic exponential smoothing model, CEEMD Verhulst
Scientific Programming 15

Table 6: Accuracy evaluation indicators of different models.

Compared models
Accuracy index
EMD-ARIMA model BP model Triple exponential smoothing model CEEMD-combined model
SSE 1.71 0.98 1.63 0.84
MAE 0.42 0.34 0.54 0.22
RMSE 0.63 0.51 1.22 0.28
MSPE 0.41 0.38 2.43 0.17
NRMSE 0.22 0.21 0.21 0.12
IA 0.56 0.79 0.45 0.82
MAPE (%) 13.46 11.09 15.33 8.36
Dstat (%) 84.78 80.45 72.32 99.52

Table 7: DM test indicators of different models.

Benchmark model
Target model
EMD-ARIMA model BP model Triple exponential smoothing model
CEEMD-combined model −2.209 (0.042) −2.184 (0.029) −2.946 (0.003)

ARIMA
0.8
0.7
0.6
CEEMD and Combined
CEEMD-ARIMA 0.71
0.5 0.55
0.4
0.3
0.2
0.23
0.1
0.55
0.45
BP CEEMD-Verhulst

0.37 0.41

EMD-ARIMA Triple Exponential Smoothing

Figure 11: Comparison of CoV of different models.

model, CEEMD-ARIMA model, and ARIMA model. In respectively, 1.37, 3.41, 2.39, and 3.52. Compared with the
terms of uncertainty quantification, the CEEMD-combi- other three methods, the CEEMD-combined model shows
nation method provides more satisfactory results (CoV) better characteristics of significant error difference and
than the other six model methods, indicating that the model avoids accidentally obtaining biased results, indicating that
has less uncertainty and can meet the actual needs. The the proposed prediction method has better stability.
stability of each model is compared by the Monte Carlo The program execution time is one of the indexes in
method. Monte Carlo sampling is a general method to modeling and calculation processing, which determines the
compare the performance of prediction methods. Its most training and prediction time of the model. The execution
remarkable feature is that it ensures the accurate comparison time of the proposed CEEMD combined model is 0.21 s
of prediction methods and avoids the biased results that may which compares with 0.33 s for EMD-ARIMA, 0.82 s for
be obtained accidentally. In this paper, the proposed model BP, and 0.41 s for the triple exponential smoothing model,
is tested by this method and the average error values of and the time taken is shorter. It shows that the proposed
different models are obtained as shown in Figure 12. It can model has a low computational time complexity and can
be seen from Figure 12 that the average error values of the achieve a faster learning process and a satisfactory and
CEEMD-combined model, the EMD-ARIMA model, the BP acceptable prediction effect, and its prediction model is
model, and the triple exponential smoothing model are, more efficient.
16 Scientific Programming

difficult to troubleshoot the faults of the system. It is difficult

Triple Exponential Smoothing
to accurately predict the failure rate. Due to the particularity
of the prediction object and the complexity of the condi-
EMD-ARIMA tions, the prediction model is not fixed. The single-term
3.52
ARIMA model not only considers the dependence of the
BP
failure rate data on the time series but also considers the
3.41 instability of random fluctuations of the data, which is
CEEMD and Combined
suitable for the prediction of aircraft failure rate. The grey
model can take advantage of the advantages of less historical
sample data and high prediction accuracy required for the
1.37 grey system prediction. At the same time, it does not need to
have typical distribution characteristics and law require-
ments for the original data. It is suitable for the prediction of
aircraft failure rate with less original information and lack of
2.39 data, but the prediction accuracy is not high in the case of
more data.

6. Conclusion
In order to provide good repair and maintenance decision
Figure 12: Average error of different models. making, establish an aircraft health management mecha-
nism, and efficiently predict the aircraft failure rate, an
aircraft failure rate prediction method based on the fusion of
According to the above overall results, it can be seen that CEEMD and combined model was established. The pre-
the CEEMD-combination model is superior to other models diction model is constructed with the failure rate of aircraft
in several evaluation indicators, and the proposed model has system failure as the original time series, through the study
better prediction performance and meets the needs of air- of performance index, uncertainty, and instability. The case
craft failure rate prediction. study results of actual aircraft failure rate data show that the
proposed prediction method can better predict the change
5. Discussion rules of aircraft failure rate and effectively improve the
prediction accuracy of aircraft failure rate. The excellent
In this study, based on the collected aircraft failure rate performance of the proposed prediction model is fully
information and related data, the information decomposition demonstrated and the prediction effect is verified. The
of the failure rate is carried out by using the complementary prediction method provides a certain idea and useful ref-
ensemble empirical mode decomposition (CEEMD) method. erence for solving the prediction problem of fault rate and
Complementary integrated empirical mode decomposition effectively improves the ability of aircraft system fault di-
(CEEMD) is a relatively novel data preprocessing method, agnosis, prediction, and comprehensive support.
which can effectively overcome the advantages of mode ali- The accuracy index of the prediction model has a SSE of
asing and white noise interference, effectively reduce the non- 0.84, a MAE of 0.22, a RMSE of 0.28, a MSPE of 0.17, and an
smoothness of time series, greatly improve the smoothness of NRMSE of 0.12. IA is 0.82 close to 1, and the mean absolute
the original sequence, and make the sequence have a percentage error (MAPE) is 8.36%. Controlling below 10%
certain regularity. Use the Hurst index to determine the indicates high accuracy, significantly narrowing the error
appropriate correlation prediction model for each de- fluctuation range and making the predicted value closer to
composition component and select the integrated moving the actual value. At the same time, the value of Dstat is
average autoregression (ARIMA) model and the grey 99.52%, and the DM test method indicates that the CEEMD-
Verhulst model as the single model of the combined combination model used to predict the aircraft failure rate is
forecasting model to carry out the combined prediction of more reasonable. At the same time, the model has less in-
group failure rate, so as to avoid the randomness and fluence of uncertainty and has the advantage of stability,
blindness of the selection of prediction model, fully con- high efficiency, and remarkable prediction effect.
sider the applicable objects of different models, and im- Although the model achieves better predictions than
prove the prediction accuracy. Fully analyze the key role of several traditional methods, there are still some challenges
different weights on the combination model and use the that require further research. (1) This study only uses his-
entropy weight method to solve the combination weight, so torical data to predict the future failure rate. In fact, the
as to avoid the impact of unified weight on the prediction aircraft system is a complex system and failures are affected
effect. Therefore, it has strong applicability. by flight time, flight sorties, ambient temperature, ambient
Due to the characteristics of complex cross-linking and humidity, maintenance capabilities, guarantee conditions,
random faults of the aircraft system, there are few data and human operations. It has a certain impact on the aircraft
sources of fault rate information samples and lack of ef- failure rate. These factors lead to strong non-linearity, un-
fective fault characteristics and diversified faults, so it is certainty, and instability in the failure rate. In the future, a
Scientific Programming 17

multi-input prediction model will be established by com- [4] J. Ma, X. Liu, X. Niu et al., “Forecasting of Landslide Dis-
bining the above influencing factors to further improve the placement Using a Probability-Scheme Combination En-
prediction performance and prediction effect. (2) It is not semble Prediction Technique,” International Journal of
perfect and systematic to evaluate the performance of the Environmental Research and Public Health, vol. 17, no. 13,
model only through several commonly used evaluation p. 4788, 2020.
indicators, and the reflected indicators and performance are [5] Z. Tian, “Approach for Short-Term Traffic Flow Prediction
Based on Empirical Mode Decomposition and Combination
single. In the follow-up, more evaluation indicators should
Model Fusion,” IEEE Transactions on Intelligent Trans-
be developed and applied to evaluate the model which can be
portation Systems, vol. 22, no. 9, pp. 5566–5576, 2021.
extended to more indicators to evaluate the model as a whole [6] H. Hu, J. Zhang, and T. Li, “A comparative study of vmd-
and improve the overall evaluation ability of the model. (3) based hybrid forecasting model for nonstationary daily
The signal decomposition method is adopted, and only the streamflow time series,” Complexity, vol. 2020, Article ID
basic CEEMD method and a basic case are used for research. 4064851, 21 pages, 2020.
In order to make the research more convincing, in the [7] Z. Tian, “Short-term wind speed prediction based on LMD
follow-up research and application, the prediction algorithm and improved FA optimized combined kernel function
research based on multiple combination models (EMD LSSVM,” Engineering Applications of Artificial Intelligence,
combination model, EEMD combination model, VMD vol. 91, Article ID 103573, 2020.
combination model, and CEEMDAN combination model) [8] Z. Tian, “Modes decomposition forecasting approach for
can be carried out to gradually improve the prefailure rate ultra-short-term wind speed,” Applied Soft Computing,
measurement accuracy and the prediction ability and uni- vol. 105, Article ID 107303, 2021.
versality of the model. At the same time, the prediction [9] Z. Tian, H. Li, and F. Li, “A combination forecasting model of
method can be applied to the prediction of aircraft failure wind speed based on decomposition,” Energy Reports, vol. 7,
pp. 1217–1233, 2021.
time, remaining life, and safety monitoring to ensure the
[10] Z. Tian and H. Chen, “Multi-step short-term wind speed
safety and reliability of the aircraft and guarantee the flight
prediction based on integrated multi-model fusion,” Applied
safety of the aircraft. Energy, vol. 298, p. 117248, Article ID 117248, 2021.
[11] Z. Tian and H. Chen, “A novel decomposition-ensemble
Data Availability prediction model for ultra-short-term wind speed,” Energy
Conversion and Management, vol. 248, p. 114775, Article ID
The maintenance data used to support the findings of this 114775, 2021.
study have not been made available because sharing the data [12] Z. Tian, S. Li, and Y. Wang, “A prediction approach using
might compromise data privacy. Moreover, the authors are ensemble empirical mode decomposition-permutation en-
not allowed to share these data due to security concerns. tropy and regularized extreme learning machine for short-
term wind speed,” Wind Energy, vol. 23, no. 2, pp. 177–206,
Conflicts of Interest 2020.
[13] Z. Tian, “Preliminary research of chaotic characteristics and
The authors declare that they have no conflicts of interest. prediction of short-term wind speed time series,” Interna-
tional Journal of Bifurcation and Chaos, vol. 30, no. 12, Article
Acknowledgments ID 2050176, 2020.
[14] N. D. Bokde, Z. M. Yaseen, and G. B. Andersen, “ForecastTB-
This study was supported by the Doctoral Startup Foun- An R Package as a Test-Bench for Time Series Forecasting-
dation of Shenyang Aerospace University (grant no. Application of Wind Speed and Solar Radiation Modeling,”
19YB30) and Scientific Research Funding Project of the Energies, vol. 13, no. 10, p. 2578, 2020.
Education Department of Liaoning Province (General [15] N. Bokde, A. Feijóo, D. Villanueva, and K. Kulat, “A review on
Program) (grant no. LJKZ0202). hybrid empirical mode decomposition models for wind speed
and wind power prediction,” Energies, vol. 12, no. 2, p. 254,
2019.
References [16] Z. Qian, Y. Pei, H. Zareipour, and N. Chen, “A review and
[1] J. Ma, Y. Wang, X. Niu, S. Jiang, and Z. Liu, “A comparative discussion of decomposition-based hybrid models for wind
study of mutual information-based input variable selection energy forecasting applications,” Applied Energy, vol. 235,
strategies for the displacement prediction of seepage-driven pp. 939–953, 2019.
landslides using optimized support vector regression,” Sto- [17] A. Altay, O. Ozkan, and G. Kayakutlu, “Prediction of aircraft
chastic Environmental Research and Risk Assessment, vol. 36, failure times using artificial neural networks and genetic al-
pp. 1–21, 2022. gorithms,” Journal of Aircraft, vol. 51, no. 1, pp. 47–53, 2014.
[2] Y. Wang, H. Tang, J. Huang, T. Wen, J. Ma, and J. Zhang, “A [18] Z. Zhao, B. Bin Liang, X. Wang, and W. Lu, “Remaining useful
comparative study of different machine learning methods for life prediction of aircraft engine based on degradation pattern
reservoir landslide displacement prediction,” Engineering learning,” Reliability Engineering & System Safety, vol. 164,
Geology, vol. 298, Article ID 106544, 2022. pp. 74–83, 2017.
[3] J. Ma, X. Niu, H. Tang, Y. Wang, T. Wen, and J. Zhang, [19] M. Kutyłowska, “Neural network approach for failure rate
“Displacement prediction of a complex landslide in the three prediction,” Engineering Failure Analysis, vol. 47, pp. 41–48,
gorges reservoir area (china) using a hybrid computational 2015.
intelligence approach,” Complexity, vol. 2020, Article ID [20] J. Z. Sikorska, M. Hodkiewicz, and L. Ma, “Prognostic
2624547, 15 pages, 2020. modelling options for remaining useful life estimation by
18 Scientific Programming

industry,” Mechanical Systems and Signal Processing, vol. 25, [36] T. Gu, J. Guo, and Z. Li, “Airborne equipment fault proba-
no. 5, pp. 1803–1836, 2011. bility prediction based on interpolation-fitting-transfer
[21] A. Lorton, M. Fouladirad, and A. Grall, “A methodology for learning algorithm,” Systems Engineering and Electronincs,
probabilistic model-based prognosis,” European Journal of vol. 40, no. 01, pp. 114–118, 2018.
Operational Research, vol. 225, no. 3, pp. 443–454, 2013. [37] Y. Zhang, Q. Pan, and H. Zhang, “New synthetic prediction
[22] L. Zhang, X. li, J. Yu, and J. Wan, “Fault prognostic algorithm method based on SVR and its application,” Journal of Air
based on relevance vector machine regresiom,” Systems En- Force Engneerng University(Natural Science Edition), vol. 184,
gineering and Electronincs, vol. 32, no. 07, pp. 1540–1543, no. 03, pp. 19–21, 2005.
2010. [38] W. Yan and J.-H. Zhou, “Predictive modeling of aircraft
[23] A. Z. Al-Garni, A. Jamal, A. M. Ahmad, A. M. Al-Garni, and systems failure using term frequency-inverse document fre-
M. Tozan, “Neural network-based failure rate prediction for quency and random forest,” in Proceedings of the 2017 IEEE
De Havilland Dash-8 tires,” Engineering Applications of Ar- International Conference on Industrial Engineering and En-
tificial Intelligence, vol. 19, no. 6, pp. 681–691, 2006. gineering Management (IEEM), pp. 828–831, IEEE, Singapore,
[24] A. Z. Al-Garni and A. Jamal, “Artificial neural network ap- December 2017.
plication of modeling failure rate for Boeing 737 tires,” [39] D. Xu, C. Zhou, C. Guan, and X. Wang, “Research on the
Quality and Reliability Engineering International, vol. 27, equipment failure rate prediction method based on ARMA-
no. 2, pp. 209–219, 2011. BP combined model,” Fire Control and Command Control,
[25] Y. Zhang and D. Wu, “Aeroengine Fault Prediction Based on vol. 46, no. 01, pp. 83–87, 2021.
Bidirectional LSTM Neural Network,” 2020 International [40] J. Liang, J. Sun, and K. Liu, “Application of a combined
Conference on Big Data, Artificial Intelligence and Internet of forecasting model in predicting failure rate of weapon
Things Engineering (ICBAIE), vol. 21, no. 05, pp. 534–538, equipment,” Commputer & Digital Engineering, vol. 40,
2020. no. 09, pp. 39-40, 2012.
[26] J. Cheng, Z. Xia, and L. Hu, “Prognostics for aeronautic [41] W. Li, N. Hou, and X. Sun, “Method for predicting failure rate
equipments based on genetic neural network,” Journal of Air of airborne equipment based on optimal combination model,”
Force Engneerng University(Natural Science Edition), vol. 12, Mathematical Problems in Engineering, vol. 2021, Article ID
no. 01, pp. 15–19, 2011. 5199982, 20 pages, 2021.
[27] X. Zhang and H. Wang, “Sparse direct support vector ma- [42] Y. Wang, Y. Liu, L. Chong, and C. Lu, “Research on failure
chine for regression and its application to failure rate pre- prediction based on empirical ModeDecomposition and LS-
SVM,” Ship Electronic Engineering, vol. 317, no. 11,
diction,” Journal of Aerospace Power, vol. 25, no. 11,
pp. 120–124, 2020.
pp. 2594–2598, 2010.
[43] R. Wang and C. Lu, “Fault rate prediction based on EMD and
[28] Y. Yang, H. Zheng, and R. Zhang, “Prediction and analysis of
grey model,” Journal of Ordnance Equipment Engineering,
aircraft failure rate based on SARIMA model,” in Proceedings
vol. 39, no. 07, pp. 127–130, 2018.
of the 2017 2nd IEEE International Conference on Compu-
[44] T. Xu, C. Lu, H. Wang, and X. Han, “Failure rate prediction
tational Intelligence and Applications (ICCIA), pp. 567–571,
method based on relevance vector EMD and GMDH re-
IEEE, Beijing, China, September 2017.
construction,” Journal of Vibration, Measurement & Diag-
[29] R. Li and R. Kang, “Research on failure rate forecasting
nosis, vol. 38, no. 06, pp. 1275–1285, 2018.
method based on ARMA model,” Systems Engineering and
[45] J.-R. Yeh, J.-S. Shieh, and N. E. Huang, “Complementary
Electronics, vol. 30, no. 08, pp. 1588–1591, 2008. ensemble empirical mode decomposition: a novel noise en-
[30] Y. Zhang, Y. Peng, P. Wang, L. Wang, S. Wang, and H. Liao, hanced data analysis method,” Advances in Adaptive Data
“Aircraft APU failure rate prediction based on improved Analysis, vol. 02, no. 02, pp. 135–156, 2010.
Weibull-based GRP,” in Proceedings of the 2017 Prognostics [46] D. Fan, L. Ni, X. Jiang, S. Fang, H. Wu, and X. Zhang,
and System Health Management Conference (PHM-Harbin), “Spatiotemporal analysis of vegetation changes along the belt
pp. 1–6, IEEE, Harbin, China, July 2017. and road initiative region from 1982 to 2015,” IEEE Access,
[31] L. Li, S. Tang, and J. Xie, “Failure prediction on the Cessna vol. 8, pp. 122579–122588, 2020.
plane based on grey theory,” Control Engineering China, [47] S. Barak and S. S. Sadegh, “Forecasting energy consumption
vol. 24, no. 07, pp. 1342–1346, 2017. using ensemble ARIMA-ANFIS hybrid algorithm,” Interna-
[32] H. Zhang, X. Cao, S. Sun, and J. Wang, “Non-equal-time- tional Journal of Electrical Power & Energy Systems, vol. 82,
interval modified Verhulst model and ist application in he- pp. 92–104, 2016.
licopters failure rates prediction,” Helicopter Technique, vol. 4, [48] S. Liu, Y. Y. Dang, and L. Wu, Grey System Theory and its
no. 04, pp. 8–11, 2021. Application, Vol. 12, Science Press, Beijing, 2017.
[33] Y. Yan-ming Yang, H. Hui Yu, and Z. Zhi Sun, “Aircraft [49] Y. Zhang, “Improved grey derivative of grey Verhulst model
failure rate forecasting method based on Holt-Winters sea- and its application,” International journal of computer science
sonal model,” in Proceedings of the 2017 IEEE 2nd Interna- Issues (IJCSI), vol. 9, 2012.
tional Conference on Cloud Computing and Big Data Analysis [50] S. Huang, B. Ming, Q. Huang, G. Leng, and B. Hou, “A case
(ICCCBDA), pp. 520–524, IEEE, Chengdu, China, April 2017. study on a combination NDVI forecasting model based on the
[34] Q. Wang and H. Yuan, “Failure rate prediction based on AR entropy weight method,” Water Resources Management,
model and residual correction,” in Proceedings of the 2017 vol. 31, no. 11, pp. 3667–3681, 2017.
Second International Conference on Reliability Systems Engi- [51] A. Botchkarev, “Evaluating performance of regression ma-
neering (ICRSE), pp. 1–5, IEEE, Beijing, China, July 2017. chine learning models using multiple error metrics in azure
[35] K. Celikmih, O. Inan, and H. Uguz, “Failure prediction of machine learning studio,” SSRN Electronic Journal, vol. 16,
aircraft equipment using machine learning with a hybrid data 2018.
preparation method,” Scientific Programming, vol. 2020, [52] D. Pradeepkumar and V. Ravi, “Forecasting financial time
Article ID 8616039, 10 pages, 2020. series volatility using particle swarm optimization trained
Scientific Programming 19

quantile regression neural network,” Applied Soft Computing,

vol. 58, pp. 35–52, 2017.
[53] H. Wang, L. Liu, S. Dong, Z. Qian, and H. Wei, “A novel work
zone short-term vehicle-type specific traffic speed prediction
model through the hybrid EMD-ARIMA framework,”
Transportmetrica B: Transport Dynamics, vol. 4, no. 3,
pp. 159–186, 2016.
[54] K. Cui and X. Jing, “Research on prediction model of geo-
technical parameters based on BP neural network,” Neural
Computing and Applications, vol. 31, no. 12, pp. 8205–8215,
2019.
[55] T. Niu, L. Zhang, B. Zhang, B. Yang, and S. Wei, “An im-
proved prediction model combining inverse exponential
smoothing and Markov chain,” Mathematical Problems in
Engineering, vol. 2020, Article ID 6210616, 11 pages, 2020.
[56] E. Uzunoglu and C. Guedes Soares, “On the model uncer-
tainty of wave induced platform motions and mooring loads
of a semisubmersible based wind turbine,” Ocean Engineering,
vol. 148, pp. 277–285, 2018.