تشخيص خطا به روش ماشين بردار پشتيبان

چكيده انگليسي :

Fault Detection By Means Of Support Vector Machine Amir Amini a amini@ec iut ac ir Date of Submission April 20 2009 Department of Electrical and computer engineering Isfahan University of Technology Isfahan 84156 83111 Iran Degree M Sc Language FarsiSupervisors Javad Askari j askari@cc iut ac ir Farid Sheikholeslam sheikh@cc iut ac irAbstract Two main problems concerned with faulty systems with high risk are complexity and improvementof the behavior of the systems For instance in aviation control systems or in chemical and nuclearplants the behavior of the system is very important whenever a fault occurs or a part of systemmalfunctions In such systems if the control part of the system doesn t detect the fault in time anddoesn t reconfigure itself suitably it s human operators must hurt and the important information andcomponents can be lost Consequently interest of researchers has grown significantly in the field of faultdetection isolation and reconfiguration Due to regulatory concerns in recent financial crises financial intermediaries credit risk assessmentis an area of renewed interest in the business community In this research we propose a new fuzzysupport vector machine to discriminate good creditors from bad ones Because in credit scoring areas weusually cannot label one customer as absolutely good who is sure to repay in time or absolutely bad whowill default certainly our new fuzzy support vector machine treats every sample as both positive andnegative classes but with different memberships By this way we expect the new fuzzy support vectormachine to have more generalization ability while preserving the merit of insensitive to outliers Wereformulate this kind of two group classification problem into a quadratic programming problem Empirical tests on three public datasets show that it can have better discriminatory power than thestandard support vector machine and the fuzzy support vector machine if appropriate kernel andmembership generation method are chosen Support vector machine is based on vapnik statistical learning theory it is one of the best algorithmsfor fault detection and isolation in dynamic systems Signature authentication of applications of asupport vector machine The purpose of signature verification is distinguish genuine signatures fromforgeries Extended regression gives a better criterion in comparison with dynamic time warping andeuclidean distance for signatures similarity Using all point matching for equalizing signal lengthdecrease differ genuine signatures from forgeries In this research a technique to differential between genuine and forgery signature based onextremum matching for equalizing signal length is proposed In addition in three tank benchmark faultis detected by using support vector machine Also it is shown than support vector machine has betterperformance than radial base function and multi layer perceptron neural networks in this application Furthermore influence of roll parameter in linear and nonlinear support vector machine is discussed infault detection and classification Key words support vector machine statistical learning theory dynamic time warping extrimum matching

اميني، امير

تشخيص خطا به روش ماشين بردار پشتيبان

تئوري يادگيري آماري , پيچش زماني پويا , تطابق نقاط اكسترمم