Anda di halaman 1dari 39

Application of Firefly & Differential evolution algorithm in prediction of time series data

Presented by Jit Saha (M.Tech C.S.E,3rd semester) Roll No:10204111005 Registration No:111020410005
under the guidance of

Satyendra Nath Mandal


Assistant Professor, Department of Information Technology Kalyani Government Engineering College, Kalyani, Nadia-741235, WB

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Abstract
Time series prediction use past to predict future Several soft-computing model have been used in this problem A new method has been proposed for prediction Firefly & DE has been separately used to this model Performance comparison with other method

5/25/2013

Index
Firefly algorithm Differential evolution Proposed method Illustration with an example Comparison Conclusion and future work References

Application of Firefly & Differential evolution algorithm in prediction of time series data

5/25/2013

Firefly algorithm
Introduction Overview Parameters of the algorithm Pseudo code

Application of Firefly & Differential evolution algorithm in prediction of time series data

Index
4

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Firefly algorithm introduction


Xin-She Yang proposed it. Inspired by the flashing behavior of fireflies Bio inspired meta-heuristic algorithm Many multimodal, non -linear problems have been solved by this algorithm
Index
5

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Firefly algorithm overview


starts from a randomly generated initial population. executes in several iterations. every firefly compares fitness to all other fireflies. the less bright firefly will move towards the brighter firefly.

Index
6

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Parameters of Firefly algorithm

Index
7

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Parameters of Firefly algorithm Contd.

Index
8

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Parameters of Firefly algorithm Contd.

Index
9

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Firefly algorithm pseudo code


Step 1: Initialize the search space & parameters. Step 2: calculate initial fitness Step 3: perform fitness comparison 3.1. lower will move towards higher in ndimension 3.2. best will move randomly in n-dimension. Step 4: if termination condition satisfied return best solution, otherwise go to step 3
Index
10

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Differential evolution algorithm


Introduction Overview Basic steps pseudo code

Index
11

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Differential evolution algorithm introduction


Proposed by price & storn [1997] Belongs to family of evolutionary algorithm Superior to Genetic algorithm (GA) Mimics the biological evolution.
Index
12

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Differential evolution algorithm overview


starts from a randomly generated initial population of chromosome executes in several generations or iterations chromosome undergoes through three stepsmutation , crossover, selection Child replace parent if it is better than parent

Index
13

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Differential evolution algorithm basic steps


Mutation Crossover selection

Index
14

5/25/2013

Mutation

Application of Firefly & Differential evolution algorithm in prediction of time series data

Index
15

5/25/2013

Crossover Operator

Application of Firefly & Differential evolution algorithm in prediction of time series data

Index
16

5/25/2013

Selection operator

Application of Firefly & Differential evolution algorithm in prediction of time series data

Index
17

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Differential evolution pseudo code


Step 1: Initialize the search space & parameters. Step 2: calculate initial fitness Step 3: do three steps for each chromosome 3.1. mutation 3.2. crossover 3.3. selection Step 4: if termination condition satisfied return best solution, otherwise go to step 3
Index
18

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Proposed method

Index
19

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Proposed method Contd.


MAPE is used as performance metric. MAPE= Total absolute percentage error / no of data in the time series. Absolute percentage error=| (Predicted value actual value)| / (actual value) * 100.

Index
20

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Illustration with an example


19 24 28 33

37
41 45 49 54 57 59 63 66
21

Index

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Firefly algorithm

Index
22

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Firefly algorithm contd.

Index
23

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Firefly algorithm contd.

Index
24

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Firefly algorithm contd.


Step 5: fitness comparison is performed between fireflies according to their fitness value
Lower will move towards higher using equation of movement Best will move randomly

Step 6: return firefly with best-fitness if termination condition is satisfied , otherwise go to step 5

Index
25

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Firefly algorithm (FA) contd.


Actual Value Predicted Value (FA) Predicted Error (%) (FA) 19 24 28 33 37 41 45 49 54 57 59 63 66 19.035626 24.007993 27.987667 32.998992 37.004838 40.999601 45.000060 49.008804 53.989571 57.000221 59.026117 63.023108 65.997185 0.187507 0.033304 0.044046 0.003054 0.013076 0.000973 0.000134 0.017968 0.019313 0.000388 0.044265 0.036680 0.004265
26

Average percentage error=0.03%


Index

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Differential evolution contd.

Index
27

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Differential evolution contd.

Index
28

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Differential evolution contd.


Step 5:return chromosome with best fitness if termination condition satisfied otherwise go to step 4.

Index
29

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Implementation example by Differential evolution (DE)contd.


Actual Value Predicted Value (DE) Predicted Error (%) (DE) 19 24 28 33 37 41 45 49 54 57 59 63 66 18.970824 24.000029 27.998296 33.006640 37.000513 40.983606 45.091680 48.909724 54.005203 57.052643 59.140763 63.001460 66.008411 0.153557 0.000121 0.006087 0.020123 0.001388 0.039985 0.203734 0.184237 0.009635 0.092357 0.238581 0.002317 0.012743
30

Average percentage error=0.07%

Index

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Comparison
Data set Method used in journal
ANN Fuzzy recurrent neural network

Error in journal method


4.73% 0.227720%

Error using FA
0.01087% 3.759390%

Error using DE
0.135254% 0.036939%

Rainfall dataset Enrollments of University of Alabama Dollar/Euro Exchange Rate Initial shoot length of Ken type mustard plant Tourism Demand in Taiwan

Chaos Theory Fuzzy soft computing model

0.487790% 0.8527%

1.177127 % 0.031152%

2.355961% 0.07422%

ARIMA

0.28%

0.003177 %

0.068899%

31

Index

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Comparison Contd.
Data set Method used in journal
FNN

Error in journal method


41.555765%

Error using FA
2.029943%

Error using DE
2.098787%

Education capital in China Absorption of cadmium by hematit Real estate prices of detached houses Rock Mass Rating dataset Population dataset

adapted neural fuzzy model

1.128380%

0.021642%

0.261729%

FNN

0.894838%

0.000577%

0.065163%

Fuzzy logic

11.20%

0.033125%

0.184286%

Fuzzy Time Series

0.13%

0.000140%

0.035973%

32

Index

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Observation & analysis


most of the cases firefly algorithm is superior to DE In very dense dataset two algorithms perform almost same In vey sparse data set DE has outperformed over firefly algorithm. Floating point calculation and time complexity of firefly algorithm is greater than DE.
Index
33

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Observation & analysis Contd.


Exhaustive exploration of search space
Fitness comparison between every two firefly helps FA for exploring search space efficiently.

Take care of local & global optima during searching


Intelligent movement between fireflies

Special emphasis on finding global optima


Best firefly also try to find out global optima

Index
34

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Conclusion & Future work


1. The proposed will be applied in other time series data to explore its limitation and weakness. 2. An elaborate parameters study is needed for the performance analysis. 3. Improve the good utilization of CPU and exploring the search space a parallel version of this may be proposed in future.
Index
35

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Reference
[1] X. S. Yang, Nature-Inspired Metaheuristic Algorithms, Luniver Press, 2008. [2]R. Storn and K. V. Price, Differential Evolution a simple and efficient heuristic for global optimization over continuous spaces, Journal of Global Optimization, vol.11, No.4, pp.341359, 1997. [3] AdilYousif, Abdul Hanan Abdullah, Sulaiman Mohd Nor and Adil Ali Abdelaziz,Scheduling Jobs On Grid Computing Using Firefly Algorithm, Journal of Theoretical and Applied Information Technology,Vol.33, No.2, pp.155-164, Nov.2011.
Index 36

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Reference Contd.
[4] Olympia Roeva, "Optimization of E. coli Cultivation Model Parameters Using Firefly Algorithm", International Journal Bioautomation, Vol.16, pp.23-32, April.2012. [5]Yudong Zhang and Lenan Wu, A Novel Method for Rigid Image Registration based on Firefly Algorithm, International Journal of Research and Reviews in Soft and Intelligent Computing, Vol. 2, No.2, pp.141-146, June.2012. [6] Puspanjali Mohapatra and Alok Raj, Indian Stock Market Prediction Using Differential Evolutionary Neural Network Model, International Journal of Electronics Communication and Computer Technology, Vol.2, Issue.4, pp.159-166, Jul.2012 Index
37

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Reference Contd.
[7] Nasimul Noman and Hitoshi Iba, Accelerating Differential Evolution Using an Adaptive Local Search, IEEE Transactions on Evolutionary Computation, Vol.12, No.1, pp.107-125, Feb.2008. [8] Ching-Wei Chien, Zhan-Rong Hsu and Wei-Ping Lee, Improving the Performance of Differential Evolution Algorithm with Modified Mutation Factor , International Proceedings of Computer Science and Information Technology, Vol.3, pp.64-69, 2011.

Index
38

5/25/2013

Application of Firefly & Differential evolution algorithm in prediction of time series data

Reference Contd.
[9] Swagatam Das, AjithAbraham,Uday K. Chakraborty, and AmitKonar, Differential Evolution using a NeighborhoodBased Mutation operator, IEEE Transactions On Evolutionary Computing, Vol.13, No.3, June.2009 [10]Cheng-Jian Lin, Chi-Feng Wu and Chi-Yung Lee, Design Of a Recurrent Functional Neural Fuzzy Network Using Modified Differential Evolution, International Journal of Innovative Computing, Information and Control,Vol.7, No.2, pp.669-683, Feb.2011.
Index
39

Anda mungkin juga menyukai