Interval Type 2 Fuzzy Logic System: Construction and Applications

Transcription

1 Interval Type 2 Fuzzy Logic System: Construction and Applications Phayung Meesad Faculty of Information Technology King Mongkut s University of Technology North Bangkok (KMUTNB) 5/10/2016 P. Meesad, JSCI2016, 28 APR

2 Agenda Introduction Background and Related Works Problem Statements The Proposed Framework Constructing IT2FLS Experimental Results Conclusion & Future Work 5/10/2016 P. Meesad, JSCI2016, 28 APR

3 Introduction Lofti A Zadeh introduced fuzzy logic in /10/2016 P. Meesad, JSCI2016, 28 APR

4 Background and related work Computational intelligence has been wildly used in Pattern classification, Regression, & Control systems. One of the prominent intelligent systems is fuzzy system. Type 1 Fuzzy set does not model well on uncertainty. General Type 2 fuzzy set was introduced in General type 2 fuzzy systems are too complicated for small hardware. Interval type 2 fuzzy systems are easier for implementation. Many works has been proposed to construct type 1 fuzzy systems. Type 2 fuzzy systems still need to study more. 5/10/2016 P. Meesad, JSCI2016, 28 APR

5 Type 1 and General Type 2 Fuzzy Systems 1 ( x m) A( xm ;, ) exp ( x m) A( xm ;, ) exp Lofti A. Zadeh 0 m 0 m l m R 5/10/2016 P. Meesad, JSCI2016, 28 APR

6 Type 2 Fuzzy Logic System Type-2 FLS Output Processing Rules Defuzzifier Crisp Output y x Crisp Input Fuzzifier Type-Reducer Type-Reduced Set (Type-1) Fuzzy Input Sets Inference Fuzzy Output Sets 5/10/2016 P. Meesad, JSCI2016, 28 APR

7 Problem Statements Grid partitioning method Too many fuzzy rules. Clustering techniques can be used but mostly need to identify the number of rules. Complicated type reduction from type 2 to type 1 fuzzy set. The reduction procedures takes too long in finding left and right points. This work proposes: 1) modification of interval type 2 fuzzy logic system 2) hybrid intelligent learning to create interval type 2 fuzzy logic system and to optimize fuzzy parameters. 5/10/2016 P. Meesad, JSCI2016, 28 APR

8 The Proposed Framework of IT2FLS Fuzzification Layer t-norm Layer Normalized Layer Inference Layer Output Layer N y 1 x 1 N y 2 Upper value N y 3 y AVE x 1 x N x M Lower value N y L 5/10/2016, x M P. Meesad, An Interval Type-2 Fuzzy System with Hybrid Intelligent Learning, WICT 2014, Dec 08-11, Mallaca, Malaysia, 2014., x 1 8

9 Constructing IT2FLS Start Initializing parameters input or rule antecedent parameters x1 x M rule consequent parameters 1 y L y Training Data One-pass Online Clustering Learning all data? abcde,,,, abcde,,,, Start Cluster-to-Rule Mapping Fuzzy Parameters Preparation Mate Selection Training Data Tuning Interval type 2 Fuzzy Sets by Hybrid Learning between GA and Steepest Descent Population Generation Fitness Evaluation Crossover Mutation Obtaining optimal solution? Fitness Evaluation Retrieving Interval type 2 fuzzy system Satisfy Solution? P. Meesad, An Interval Type-2 Fuzzy System with Hybrid Intelligent Learning, WICT 2014, Dec 08-11, 5/10/2016 Stop 9 Mallaca, Malaysia, Stop

10 Incremental Fuzzy Neural Network (ILFN) x 2 Class 1 Class 2 0 x 1 5/10/2016 P. Meesad, JSCI2016, 28 APR

11 Cluster to Rule Mapping x 2 low medium high Cluster by Using ILFN 5/10/ low high P. Meesad, An Interval Type-2 Fuzzy System with Hybrid Intelligent Learning, WICT 2014, Dec 08-11, 0 x Mallaca, Malaysia,

12 ILFN2RULE Algorithm w P1 w PM w T low medium high 1 Linguistic Label {1: low, 2: medium, 3: high} Knowledge Base Antecedent Consequent w TM Direct Mapping 5/10/2016 P. Meesad, JSCI2016, 28 APR

13 Experimental Results Summary Data Data Set Attribute Type Number of Records (used) Number of Features (used) Bank Market Real (4119) 17 (7) Banknote Authen Real Car Evaluation Categorical Wilt Real /10/2016 P. Meesad, JSCI2016, 28 APR

14 Experimental Results Data Set % accuracy of each technique C4.5 MLP SVM ANFIS IT2FIS Bank Market Banknote Authen Car Evaluation Wilt Data /10/2016 P. Meesad, JSCI2016, 28 APR

15 Mackey Glass Time Series 5/10/2016 P. Meesad, JSCI2016, 28 APR

16 Stock Exchange of Thailand 5/10/2016 P. Meesad, JSCI2016, 28 APR

17 Conclusion Modification of Interval Type-2 Fuzzy Logic System is proposed. A method to create and optimize Interval Type-2 Fuzzy Logic System is proposed. Online one-pass clustering is first performed; incremental learning fuzzy neural network (ILFN) is used. Then clusters are mapped to rules, including all fuzzy parameters. The fuzzy parameters are then optimized by hybrid learning genetic algorithm and steepest decent. Experimental results showed that the proposed technique is comparable to existing works. For future work, hardware implementation in real time on FPGA for control system. 5/10/2016 P. Meesad, JSCI2016, 28 APR

18 Thank you for your attention Questions or Suggestions? 5/10/2016 P. Meesad, JSCI2016, 28 APR