Forecasting Chaotic Time Series

International Journal of Hybrid Information Technology
Vol.6, No.6 (2013), pp.369-380

http://dx.doi.org/10.14257/ijhit.2013.6.6.33
Forecasting Chaotic Time Series with Wavelet Neural Network

Improved by Particle Swarm Optimization
Hui Li1, 2, Dechang Pi1 and Min Jiang1
1
College of Computer Science and Technology, Nanjing University of Aeronautics

and Astronautics
29 Yudao Street, Baixia District, NanJing
2
Department of Information Technology, Jinling Institute of Technology
99 Hongjing Ave., NanJing
1
lihui@jit.edu.cn, 2dc.pi@nuaa.edu.cn , 3alicehh_ok@126.com

Abstract
The prediction of chaotic time series is an important research issue. To improve the
prediction accuracy, a hybrid approach called WNN-PSO is proposed, which based on the
self-learning ability of wavelet neural network, whose parameters are optimized by particle
swarm optimization. The WNN-PSO method has higher prediction accuracy, fast
convergence, and heightens the ability of jumping the local optimums. The experiment results
of the prediction for chaotic time series show the feasibility and effectiveness of the proposed
method. Compared with wavelet neural network and BP neural network, the proposed method
are superior to them. Finally, the WNN-PSO is applied to predict the life energy consumption
of china in our lives.
Keywords: Chaotic time series, Wavelet neural network, Particle swarm optimization,
Forecast
1. Introduction
Chaos seems an intrinsic random feature of deterministic system, and exists everywhere in
our daily lives. It is not an individual accidental phenomenon. The dynamic system of chaos
possesses fixed characteristics and rules. If we can find the determination rules, it is possible
to predict the performance of the non-linear dynamic system in the future [1]. Chaotic time
series is an important branch of chaos theory. It has widely applied in various fields of natural
science and social science, such as, hydrological forecasts, stock market, sunspots, image
processing, which has important practical value and significance. Therefore, the
reconstruction model and prediction of chaotic time series are an important research issue.
The classical chaotic time series forecasting methods are global prediction approach [2],
local prediction method [3], adaptive prediction [4], the largest lyapunov exponents
prediction [5], and so on. Global prediction approach is clear in concept, but it is difficult
with larger noise. Local prediction method found the adjacent points of prediction points in
phase space, and took the next point of the most nearest point on the track as the prediction
output. Once the attractor overstepped the corresponding area, the model would fail with
poorer accuracy. The adaptive prediction had a higher demand for tracking identification and
real-time recursive capabilities. The largest lyapunov exponents were not an intelligent
technology and needed to calculate LEs for predicting chaotic time series data. Scholars at
ISSN: 1738-9968 IJHIT

Copyright 2013 SERSC

Vol.6, No.6 (2013)
home and abroad had also proposed a variety of intelligent prediction algorithms, such as
chaotic prediction based on neural networks [6, 7, 8], gray theory prediction method [9],
fuzzy neural network [10], time series prediction method based on the reserve pool [11],
support vector machine method [12], wavelet neural network prediction method [13, 14].
These approaches have been applied in some fields, but with poor accuracy rate or slow speed
because of single algorithm.
Recently, in order to improve the accuracy and speed of time series prediction, some
combined forecasting algorithms are proposed. WANG [15] et al., gave an approach of BP
neural network based on improved particle swarm optimization to heighten the ability of
jumping the local optimums because of introducing crossover operator and mutation operator.
LI Xiang et al., [16] presented an improved wavelet neural network combined with PSO
algorithm to reduce the curse of dimensionality and to increase the precision. Meng Yuebo et
al., [17] provided a WNN-IPSO algorithm to improve the modeling capability of WNN.
Georgios Sermpinis et al., [18] proposed hybrid neural network architecture of particle swarm
optimization and Adaptive Radial Basis Function (ARBF-PSO) to forecast foreign exchange
rates. These methods were used for predicting time series. Arash Miranian and Majid
Abdollahzade [19] introduced a local neuro-fuzzy approach based on the LSSVMS algorithm.
It is powerful in modeling and predicting for nonlinear and chaotic time series, but it was a
local modeling method, which was a little difficult.
Artificial Neural Network (ANN) was put forward based on the research results of modern
neuroscience, which had already applied in a wide range of fields. The wavelet neural
network (WNN) is a combination of wavelet analysis theory and ANN, which takes full
advantage of time-frequency characteristics of wavelet transform and self-learning ability of
the neural network. Compared with ANN, WNN takes the wavelet basis function as the
transfer function of the hidden layer, with the signal forward propagation, and deviation back
propagation, which shows better prediction accuracy, convergence rate and fault tolerance to
the complex nonlinear, uncertain, unknown system [20, 21]. In addition, WNN has the
advantages of adaptive learning ability, strict theory bases and high nonlinear mapping
capability. The activation function of WNN introduces translation factor, dilation factor and
other parameters, which makes the network have a better predictive ability and fault tolerance
than other networks. Due to fewer research on using particle swarm optimization (PSO)
algorithm to establish the chaotic time series prediction model, we introduce PSO algorithm
into the field of prediction modeling to train the WNN for better performance, which has an
important significance in theory and practice.
In this paper, a hybrid model based on WNN and PSO is presented. The proposed method
has some features: (1) it owns better prediction accuracy than single existed algorithm; (2) it
leads fast convergence and global optimum in our study; (3) it can improve the learning
ability of WNN. The main idea of the proposed method is as follows: firstly, PSO was used to
tune the parameters of WNN for satisfying the requirements of chaotic time series, and WNN
was trained to learn and extract the features of chaotic time series. Then, the hybrid model
was built to predict the time series data. Finally, simulation results showed the effectiveness
and efficiency of the proposed method. In addition, the model was applied to predict life
energy consumption in china.
The rest of the paper is organized as follows. The hybrid prediction model for chaotic time
series is presented in Section 2. Section 3 discusses the experimental results for the proposed
algorithm. Conclusions are given in Section 4.
370

Vol.6, No.6 (2013)
2. Hybrid Model of WNN and PSO

2.1. Particle swarm optimization
Where there is great love, there are always miracles. Love is like a butterfly. It goes where
it pleases and it pleases where it goes. If I had a single flower for every time I think about
you, I could walk forever in my garden. Within you I lose myself, without you I find myself
wanting to be lost again. At the touch of love everyone becomes a poet.
Particle Swarm optimization (PSO) [21] is a typical swarm intelligence algorithm, which is
used to seek the optimal solution of the optimization problem. It stems from the simulation of
bird predation. When the birds prey, the most effective and simplest way of each bird finding
food is to search around the location of the bird closest to the food. PSO algorithm first
assumes that every possible solution is a particle in the search space; the features of each
particle are expressed as position vector X , velocity V , and fitness value f . Among them, the
particle's position means a temporary solution in the solution space; the velocity represents
the direction and distance of particles motion. In order to get the best value in the solution
space, they will be adjusted accordingly by the particle itself and other particles.
Generally, in the process of search, a particle updates itself by following two extreme
values in the solution. One extreme value named Pb , is the optimal solution gained by the
particle itself, and is the best fitness value calculated by the fitness function. The other one
named Pg , is the optimal solution found by all particles at the present, and the best fitness
value of all particles in the solution space. During the updating process, if the position of a
particle is moved, the corresponding fitness value will be re-calculated; Pb and Pg of the
particle will be renewed according to the new fitness value. Fitness value of the particle is
calculated by a pre-defined fitness function, whose value indicates the relationship of the pros
and cons of the particles in the search space. After constantly moving by own experience and
its neighbors experience, the particle will eventually seek to the optimal value in the solution
space.
Assume a swarm of particles named X X , X , , X , the position and the velocity of the i th particle in D-dimensional solution space are represented as X i xi1 , xi 2 , , xiD T and
V V ,V , ,V respectively.
1
i1
i2
iD
Where, X i represents a potential solution in the search space. The best individual extreme
value of the particle i is expressed as Pbi Pbi1 , Pbi 2 , , PbiD T , and P P , P , , P presents
T
g1
g2
gD
the best position among all particles in the population X . By iterative calculation again, the
velocity and position of the particles are updated through Pb and Pg according to the
following formulas:
k 1
Vid wVid c1 r1 Pbid X id c2 r2 Pgd X id

k
X idk 1 X idk Vidk 1
(1)
(2)
Where d 1, 2, , D , D is the dimension of the space; i 1, 2, , n , n is the swarm size; k is

the times of evolutionary iteration; c1 and c2 are the acceleration coefficients which are
nonnegative constants, Here the value is 2; r1 and r2 are two generated uniformly distributed
random numbers in the range of [0,1]; w is the internal weight coefficient. In order to prevent
371

Vol.6, No.6 (2013)
the particles falling into local optimal solution, or blindly searching in the solution space, Vi
and X i of the particle are limited within a set range Vmax ,Vmax and X max , X max
respectively.
Particle's fitness value is calculated by the fitness function, whose value is good or bad
represents the pros and cons of the particles. Fitness function is defined by equation (3).
1 n
1 n
y c1 exp 0.2 x 2j exp cos(2 x j ) c1 e
n j 1
n j 1
(3)
Where c1 c2 20 , e=2.71282 , n 2 .
2.2. Framework of hybrid structure

Traditional wavelet neural network uses gradient descent method to search, which is a
typical local search algorithm. The network is easy to fall into local minima in the training
process. The initial parameters of WNN are assigned based on prior knowledge, which is
subjective and cannot reach a good accuracy of prediction. In PSO algorithm, a particle
represents a possible solution during the optimization iteration. In the literation process, the
formula is simple; the calculation speed of the method is much faster than the gradient
descent. In this paper, the PSO algorithm is used to optimize the structural parameters of the
wavelet neural network, which does not need to get the derivative and differential of the
activation function. Simultaneously, the parameters are adjusted through the iterative formula
easily, so the network can jump out of local extreme.
The framework of prediction model based WNN improved by PSO is shown in Figure 1.
Firstly, the WNN model is used to build the initial network topology. Then, PSO algorithm is
employed to search for the optimal particle in the solution space, which is the optimal
solution, and the initial parameters of wavelet neural network are assigned. Thirdly, the
predicted data is decomposed into two parts. The characteristics of each part of the data are
analyzed, and then train the network. Finally, the data are forecasted using the proposed
model. Compared the predicted results with the actual data, the performance of the network is
analyzed.
start
WNN
modeling
PSO
optimization
end
WNN
predicting
WNN training
Figure 1. The prediction model of WNN improved by PSO

During the modeling process, the optimal structure of WNN depends on the number of
nodes of different layers, especially the number of hidden layer nodes. If the number of
hidden layer nodes is too small, the training effect is not very good. If too many, the training
speed will be affected. The network topology used in the proposed model is the three-layered
372

Vol.6, No.6 (2013)
structure. The structure is 4-6-1. That is, the number of the input layer nodes is 4. The number
of hidden layer nodes is 6. The number of the output layer nodes is 1.
2.3. Wavelet neural network improved by PSO
In order to improve the accuracy of the prediction, PSO algorithm is used to optimize the
weights and the wavelet coefficients of wavelet neural network (WNN-PSO). The network
structure parameters of WNN are expressed as the position vector of the particle X , and are
tuned through the iteration of equation (1) and (2) of PSO to find the optimal solution, where
the fitness of the particle is calculated by equation (3). The process is repeated many times
until it reached the pre-defined accuracy. The training process flowchart of PSO-WNN is
shown in Figure 2.
Define position
vector X
Initialize
parameters of
particle swarm
Evaluate
fitness value of
each particle
Update Pb,Pg with

optimal value
Renew the
particle state
start
no
end
Is in optimal
state?
no
Output
WNN
predicting
yes
< ?
yes
Train WNN
network
Initialize WNN
parameters
Figure 2. The flow chart for learning algorithm in WNN with PSO
The specific process of proposed model is described as follows:
Step 1: Determine the dimension of position vector X and the velocity vector V of each
particle, Dimension D = number of hidden layer nodes * number of the input layer nodes +
number of output layer nodes * number of hidden layer nodes + number of translation
parameters + number of dilation parameters, namely, D I * J J * K 2 * J .
Step2: Initialize the population, ensure the structure parameters of the network, and
generate the initial particle swarm.
Step 3: Calculate the fitness value of each particle in particle swarm.
Step 4: Compare the fitness value of each particle in particle swarm with Pb , if the fitness is
better, then update Pb .
Step 5: Compare the fitness value of each particle in particle swarm with Pg , if the fitness
is better, then update Pg .
Step 6: Update the velocity V and position X of the particle in accordance with the formula
(1) and (2).
373

Vol.6, No.6 (2013)
Step 7: If the particle reaches a good position or the maximum number of iterations, the
calculation process will be halted. Otherwise, go to step 3, continue to find the optimal
values.
Step 8: Initialize the parameters of WNN utilizing the optimal solution, and train the
network. Then, predict the data. Finally, output the forecast results.
During the entire optimization process, the velocity and position of each dimension of the
particle cannot exceed the set interval of the algorithm, which are in the range
V , V and X , X respectively. The range can be set according to the actual
max
max
max
max
needs of different network parameters. In the iterative process, reducing Vmax can
improve the local search ability of particles, and then jump out of local optimum.
3. Empirical Results
The proposed mixed model WNN-PSO is used to predict chaotic time series data.
Compared with existing models (BP, WNN), the proposed method can improve the forecast
accuracy. Finally, the model is applied to our real life to predict the life energy consumption
of china, and the prediction performance is analyzed.
In order to evaluate and analyze the experimental results, the mean square error (MSE) is
used to measure the performance of different methods. The formula is shown in equation (4).
MSE
Y Y '
n
i
(4)
i 1
Where Y and Y ' represent the actual value and predicted value of chaotic time series
respectively.
3.1. Prediction of Mackey-Glass Time Series
To validate the algorithm, the experimental data are produced by the Mackey-Glass
differential delay equation. The equation is:
dx(t )
ax(t )
x(t )
dt
1 x (t )
(5)
Choose 0.2 , 0.1 , 10 ,the formula (5) is transformed into equation (6).
dx(t ) 0.2 x(t )
0.1x(t )
dt
1 x10 (t )
(6)
The time series is the function of parameter , when 16.8 , the system shows chaotic
behavior, so let 17 . Set the initial conditions, it can get the time series x (t ) . Figure 3 shows
the chaotic time series data when 17 .
374

Vol.6, No.6 (2013)
1.6
1.4
1.2
x(t)
0.8
0.6
0.4
0.2
100
200
300
400
500
600
time(sec)
700
800
900
1000
Figure 3. Mackey-Glass time series data

The prediction results and MSE are used to analyze the performance of BP, WNN, and
WNN-PSO. The chaotic time series data used in the experiment contain 1000 time points and
the corresponding data values, which selected as the sample data. The former 800 data points
were taken as the training samples, the other 200 data points were chosen as the test data
samples. Three kinds of networks trained carried out 100 times on the average. The below
experiment figures give the distribution of predicted and desired results. Then, compared the
forecast data with the real data, the prediction accuracy of the various algorithms were
analyzed.
Firstly, BP neural network is used to predict the test data, the prediction results can predict
the trend of chaotic time series, but the larger deviation is appeared. Figure 4 shows the
prediction results and the prediction deviation.
0.1
1.4
Actural Chaotic Time Series
Predicted Chaotic Time Series
1.3
0.08
0.06
1.1
0.04
Prediction Deviation
1.2
x(t)
1
0.9
0.8
0.02
0
-0.02
0.7
-0.04
0.6
-0.06
0.5
-0.08
0.4
20
40
60
80
100 120
time(sec.)
140
160
180
(a) The prediction results of BP network
200
-0.1
20
40
60
80
100 120
time(sec.)
140
160
180
200
(b) The prediction deviation of BP network
Figure 4. The Experimental results by BP Network

Then, WNN method is used to experiment on the data. Compared with BP network, the
results has been slightly improved. Figure 5 shows the results of WNN prediction, and the
prediction error.
375

Vol.6, No.6 (2013)
0.1
1.4
0.08
1.3
0.06
1.2
0.04
1.1
x(t)
1
0.9
0.8
0.02
0
-0.02
0.7
-0.04
0.6
-0.06
0.5
-0.08
0.4
-0.1
0
20
40
60
80
100 120
time(sec.)
140
160
180
200
(a) The Experimental results of WNN
20
40
60
80
100
120
time(sec.)
140
160
180
200
(b) The Experimental deviation of WNN
Figure 5. The Experimental results by WNN network

In order to obtain a better solution, PSO is used to optimize wavelet neural network, then
WNN is trained to predict the test data. The experimental results is shown in Figure 6(a), the
prediction error is shown in Figure 6(b).
1.4
1.3
0.1
1.2
0.08
1.1
0.06
0.04
x(t)
1
0.9
0.8
0.7
0.02
0
-0.02
-0.04
0.6
-0.06
0.5
0.4
-0.08
20
40
60
80
100
120
time(sec.)
140
160
180
200
(a) The prediction results of WNN-PSO
-0.1
20
40
60
80
100
120
time(sec.)
140
160
180
200
(b) The prediction deviation of WNN-PSO
Figure 6. The Experimental Results of WNN-PSO

As can be seen from the above experimental results, the proposed WNN-PSO can bring a
higher degree of accuracy, which makes the predicted experiment results be closer to the real
data. In order to assess the prediction errors, the MSEs of three network models above are
calculated. As shown in Table 1, the WNN optimized by PSO owns the smallest prediction
deviation when predicting chaotic time series.
Table 1. The comparison of MSE of three prediction model
Network
Topology
MSE
376
BP
9.6814 10
WNN
6
5.7267 10
PSO-WNN
6
3.6908 10

Vol.6, No.6 (2013)
The prediction accuracy of WNN depends on the optimal solution of PSO algorithm,
which affected by the quantity of swarm. The experiment sets the number of swarm for 20,
30, 40 and 50, then the proposed WNN-PSO method is used to predict the data respectively.
As the population increasing, the MSE decreases. The results of the comparison are shown in
Table 2.
Table 2. The comparison of MSE with increased population in WNN-PSO
Population
MSE
20
30
3.6908 10
40
3.5626 10
50
3.4773 10
3.3964 10
3.2. Real-world application prediction of life energy

In this section, the proposed WNN-PSO model is applied to the prediction of life energy
data. The data sets used in the experiments, include the total consumption from the year 1978
to 2009 of the life energy, which comes from the National Bureau of Statistics website [23].
The data contains 32 data points.
The training data of the experiments of three methods are the life energy consumption
between the years 1978 to 2004, the testing data sets are from the year 2005 to 2009. The
number of training times for each network is 100 times, and the prediction errors of different
models are given to measure the accuracy of the prediction.
The MSE of three models are given in Table 3. As shown in Table 3, the proposed model
WNN-PSO can provide better prediction accuracy compared with other two models.
Table 3. The comparison of MSE of three prediction model
Network
Topology
MSE
BP
7.2736 10
WNN
4
4.6534 10
PSO-WNN
4
2.8921 10
4. Conclusions
The chaotic data exists in our daily lives, the prediction of chaotic time series is a research
hotspot in chaotic areas [24, 25]. Because of self-learning, fault tolerance, non-linearity of
wavelet neural network, the features of optimization and simplicity of PSO, the forecast
model based on these advantages is built for predicting chaotic time series. The proposed
model avoids the weakness of falling into local optimization easily, slow convergence, poor
prediction accuracy, which owns better global feature and convergence rate. Compared with
traditional classical models, the proposed method can further improve the accuracy of the
chaotic time series. Moreover, the achievement of the model is efficient enough to be applied
in life energy consumption data.
The research gives a short-term prediction of chaotic time series, and cannot give an
effective long-term prediction of it, which is a difficult issue in chaotic field. The long-term
forecast of chaotic time series will be the focus of further research.
377

Vol.6, No.6 (2013)
Acknowledgements
This work was supported by Qing Lan Project of Jiangsu province, 333 Project of Jiangsu
province, Aeronautical Science Foundation of China (No. 20111052010), the National 863
Program (No. 2007AA01Z404), Universities Natural Science Research Project of Jiangsu
Province (No. 13KJD520005), Modern Educational Technology in Jiangsu province (2013-R26144).
References
[1] H. Yoshito, O. Makito and A. Kazuyuki, Chaos in neurons and its application: Perspective of chaos
engineering, Chaos, vol. 22, no. 4, (2012).
[2] S. F. Su, C. B. Lin and Y. T. Hsu, A high Precision global Prediction approach based on Local prediction
approaches, IEEE Transactions on Systems, Man and Cybernetics, Part C, vol. 32, no. 4, (2002).
[3] J. D. Farmer and J. Sidorowich, Predicting chaotic time series, Physical Review Letters, vol. 59, no. 8,
(1987).
[4] J. C. Gan and X. C. Xiao, Adaptive predict-filter of chaotic time series constructed Based on the
neighbourhood in the reconstructed phase space, Acta Physica Sinica, vol. 52, no. 5, (2003).
[5] J. Zhang, K. F. Man and J. Y. Ke, Time series prediction using Lyapunov exponents in embedding phase
space, Computers & Electrical Engineering, vol. 30, no. 1, (2004).
[6] B. Sivakumar, A. W. Jayawardena and T. M. K. G. Fernando, River flow forecasting: use of phase-space
reconstruction and artificial neural networks approaches, Journal of Hydrology, vol. 265, no. 1, (2002).
[7] M. Han, J. Xi, S. Xu and F. Yin, Prediction of chaotic time series based on the recurrent predictor neural
network, IEEE Transactions on Signal Processing, vol. 52, no. 11, (2004).
[8] C. Wong and M. Versace, CARTMAP: a neural network method for automated feature selection in financial
time series forecasting, Neural Comput. & Applic., vol. 21, no. 5, (2012).
[9] E. Kayacan, B. Ulutas and O. Kaynak, Grey system theory-based models in time series prediction, Expert
Systems with Applications, vol. 37, no. 2, (2010).
[10] N. K. Kasabov and Q. Song, DENFIS: Dynamic Evolving Neural-Fuzzy Inference System and Its
Application for Time-Series Prediction, IEEE TRAN. On Fuzzy System, vol. 10, no. 2, (2002).
[11] M. Han, Z. W. Shi and W. Guo, Reservoir neural state reconstruction and chaotic time series prediction,
Acta Physica Sinica, vol. 56, (2007).
[12] B. Zhu and Y. Wei, Carbon price forecasting with a novel hybrid ARIMA and least squares support vector
machines methodology, Omega, vol. 41, (2013).
[13] Y. Chen, B. Yang and J. Dong, Time-series prediction using a local linear wavelet neural network,
Neurocomputing, vol. 69, no. 4, (2006).
[14] Z. Zainuddin and O. Pauline, Modified wavelet neural network in function approximation and its application
in prediction of time-series pollution data, Applied Soft Computing, vol. 11, no. 8, (2011).
[15] Z. Wang, X. Zhao, B. Wan and J. Xie, Research of BP Neural Network based on Improved Particle Swarm
Optimization, Journal of Networks, vol. 8, no. 4, (2013).
[16] X. Li, S. D. Yang, J. X. Qi and S. -x. Yang, Improved wavelet neural network combined with particle swarm
optimization algorithm and its application, J. Cent. South. Univ. Technol., vol. 13, no. 3, (2006).
[17] Y. B. Meng, J. H. Zou and X. S. Gan, Research on WNN aerodynamic modeling from flight data based on
improved PSO algorithm, Neurocomputing, vol. 83, no. 4, (2012).
[18] G. Sermpinis, T. Konstantinos and K. Andreas, Forecasting foreign exchange rates with adaptive neural
networks using radial-basis functions and Particle Swarm Optimization, European Journal of Operational
Research, vol. 225, no. 3, (2013).
[19] A. Miranian and M. Abdollahzade, Developing a Local Least-Squares Support Vector Machines-Based
Neuro-Fuzzy Model for Nonlinear and Chaotic Time Series Prediction, IEEE TRAN. On Neural Networks
and Learning Systems, vol. 24, no. 2, (2013).
[20] Q. H. Zhang and A. Benveniste, Wavelet network, IEEE Trans. on Neural Network, vol. 3, no. 6, (1992).
[21] Q. H. Zhang, Using wavelet network in nonparametric estimation, IEEE Trans. Neural Networks, vol. 8, no.
2, (1997).
[22] J. Kennedy and R. C. Eberhart, Particle swarm optimization, Proc. IEEE Int. Conf. Neural Networks IV,
(1995), pp. 1942-1948.
[23] http://219.235.129.58/welcome.do.
378

Vol.6, No.6 (2013)
Authors
Hui Li
She received her M.Sc. in Computer Sciences and Technology (2006)
from Northwestern Polytechnical University. Now she is a Ph.D.
candidate at the College of Computer Science and Technology, Nanjing
University of Aeronautics and Astronautics. Since 2006 she is a lecture
of Jinling Institute of Technology. Her current research interests include
different aspects of Artificial Intelligence and Data Minning.
Dechang Pi.
He received his M.Sc. in Computer Sciences and Technology (1996)
and PhD in Computer Sciences (2002) from Nanjing University of
Aeronautics and Astronautics. Now he is full professor and Ph.D.
supervisor of computer at the College of Computer Science and
Technology, Nanjing University of Aeronautics and Astronautics. His
current research interests include different aspects of Artificial
Intelligence and Data Mining.
Min Jiang
He received his M.Sc. in the College of Computer Science and
Technology (2008) from Nanjing University of Aeronautics and
Astronautics. Now he is a full research engineer in a research agency.
His current research interests include different aspects of Artificial
Intelligence and Distributed Systems.
379

Vol.6, No.6 (2013)
380

Forecasting Chaotic Time Series

Uploaded by

Copyright:

Available Formats

Forecasting Chaotic Time Series

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Forecasting Chaotic Time Series

Uploaded by

Copyright:

Available Formats

International Journal of Hybrid Information Technology

Vol.6, No.6 (2013), pp.369-380