Research On Fuzzy Support Vector Machine And Its Application On Fault Diagnosis

Support vector machine (SVM), which is one of the standard tools for machine learning and data mining, can deal with classification problems and regression problems successfully. As an immature technology, however, there is a lot of uncertain information in the objective world. If the training set on SVM contains uncertain information, then the standard model on SVM will be powerless.In this paper, through integrating fuzzy set with SVM, the conception of fuzzy membership is introduced into the least square support vector machine (LS_SVM), which overcomes the effect on the uncertain information and SVM is time-consuming to solve quadratic programming problems. Finally, the model of FLS_SVM (fuzzy least square vector machine, FLS_SVM) was introduced into fault diagnosis field.In this paper, the main work for the constructions of fuzzy membership functions is as follows:1. This part mainly has researched membership function model based on the sample space. Because the training samples affected by noise, its regression performance will be affected. Therefore, according to the degree of regression curve point away from the training samples, give each sample a different value of degree of membership, to suppress impact by noise on the support vector machine training.2. This part mainly has researched fuzzy membership function models based on the kernel space. The method described above is about defining the fuzzy membership function in the original space, when the original sample space into a high dimensional space by mapping, as they are constructed over the role when the plane is different, so degree of membership is defined by the different contribution of the nuclear space of the samples.The FLS_SVM model is built by the above two methods of determining the membership function, and finally, it is applied to the flight control system speed sensor fault diagnosis. FLS_SVM is trained out of line, and used online. After being trained, FLS_SVM is used to simulate system dynamic characteristic. The simulation result is compared with actual output, and then fault error is drawn. Taking yaw angular rate sensor fault diagnosis for example has simulated.The simulation result shows that, FLS_SVM can simulate the system more accurately, thus fault message of sensor is diagnosed in time. Experiments demonstrate the effectiveness of the method.