انعطاف پذیر رگرسیون بردار پشتیبانی و کاربرد آن در تشخیص خطا
|کد مقاله||سال انتشار||مقاله انگلیسی||ترجمه فارسی||تعداد کلمات|
|25908||2013||13 صفحه PDF||سفارش دهید||6412 کلمه|
Publisher : Elsevier - Science Direct (الزویر - ساینس دایرکت)
Journal : Acta Automatica Sinica, Volume 39, Issue 3, March 2013, Pages 272–284
Hyper-parameters, which determine the ability of learning and generalization for support vector regression (SVR), are usually fixed during training. Thus when SVR is applied to complex system modeling, this parameters-fixed strategy leaves the SVR in a dilemma of selecting rigorous or slack parameters due to complicated distributions of sample dataset. Therefore in this paper we proposed a flexible support vector regression (F-SVR) in which parameters are adaptive to sample dataset distributions during training. The method F-SVR divides the training sample dataset into several domains according to the distribution complexity, and generates a different parameter set for each domain. The efficacy of the proposed method is validated on an artificial dataset, where F-SVR yields better generalization ability than conventional SVR methods while maintaining good learning ability. Finally, we also apply F-SVR successfully to practical fault detection of a high frequency power supply.
نتیجه گیری انگلیسی
The parameter setting strategy for SVR has been stud-ied for a decade, however there is still no general consensus on theoretical estimation. This enhances the di±culty for ordinary users in applying SVR to practical applications.Moreover, as investigated in Section 1, the conventional SVR approaches are also in a dilemma of selecting rigorous or slack parameters for training some complicated distribu-tions. Hence, a °exible support vector regression (F-SVR) approach has been proposed in this paper, which is capa-ble of adapting the parameters to the sample distribution instead using of ¯xed parameters.With automatic parameter setting, the learning ability is maintained in the F-SVR whilst the number of support vec-tors is also reduced. The proposed approach o®ers a better trade-o® between over-¯tting and under-¯tting. The e®ec- tiveness of the F-SVR is validated by applying it to an arti-¯cial dataset, where appropriate hyper-parameters are hard to be set due to its complicated sample distribution. During the experiment, the F-SVR succeeded in making a regres-sion with su±cient learning and better generalization abil-ity, and showed its superiority to conventional approaches.We also made an attempt of employing the approach to de- tect the leaning excitation fault for high frequency power supply. The F-SVR o®ered a good description of the rela- tionship between the SPC and operating status, and showed its applicability in practice. However, while the setting of the three parameters is avoided, a new parameter for sample division \k" is aroused in this approach. This parameter may greatly a®ect the computational cost during training, and requires further study in the future.