Fault Diagnosis Improved Support Vector Machine Based On High-voltage Circuit Breakers

High voltage circuit breaker, as one of the most significant equipment in the electric system, plays an important role in assuring of running the electric system properly. The high voltage circuit breaker should have a real-time monitoring of its working situation, and should be overhauled and early warned, which not only assures of the electric system’s running securely but also avoids producing new problems and reducing the waste of money caused by overhauling the equipment blindly.Nowadays, state recognition methods of troubleshooting include mainly statistical recognition recognition method, which is a kind of classification algorithm that would be reasonable on condition that there are enough samples. Support vector machine, developing on the basis of statistical theory, is a new classification algorithm exclusive of less samples. With stronger theoretical evidence, it has made up the imperfections of traditional statistics, and it’s appropriate for small samples.To achieve intelligent diagnosis towards the state of high voltage circuit breaker in intelligential, specific to the insufficient sample data in studying the state of breaker failure, this paper gives a detailed analysis about working principle of support vector machine, and applies it to the state classification of high voltage circuit breaker, and shows a deep research into the influences that the choice of classification model, kernel function type, and kernel function parameters in support vector machine will have on fault classification performance. Also, this paper analyzes classified effects of both different kernel functions and classification models in details, and pays more attention to the means to optimize the radial basis kernel function parameters, in the meantime, it creates a new model of fault diagnose based on genetic algorithm optimizing kernel function parameters. According to the comparison of experimental results between modified GA-SVM algorithm one using support vector machine classifiers which adopts conventional kernel function parameter optimization methods to increase loss function parameter s and modified GA-SVM algorithm two whose every operator in genetic operations has been improved, the result of comparison shows that two modified GA-SVM algorithms have a better classified effects.