کد مقاله کد نشریه سال انتشار مقاله انگلیسی نسخه تمام متن
805649 1468251 2013 8 صفحه PDF دانلود رایگان
عنوان انگلیسی مقاله ISI
A dynamic particle filter-support vector regression method for reliability prediction
موضوعات مرتبط
مهندسی و علوم پایه سایر رشته های مهندسی مهندسی مکانیک
پیش نمایش صفحه اول مقاله
A dynamic particle filter-support vector regression method for reliability prediction
چکیده انگلیسی


• A dynamic PF–SVR method is proposed to predict the system reliability.
• The method can adjust the SVR parameters according to the change of data.
• The method is robust to the size of training data and initial parameter values.
• Some cases based on both artificial and real data are studied.
• PF–SVR shows superior prediction performance over standard SVR.

Support vector regression (SVR) has been applied to time series prediction and some works have demonstrated the feasibility of its use to forecast system reliability. For accuracy of reliability forecasting, the selection of SVR's parameters is important. The existing research works on SVR's parameters selection divide the example dataset into training and test subsets, and tune the parameters on the training data. However, these fixed parameters can lead to poor prediction capabilities if the data of the test subset differ significantly from those of training. Differently, the novel method proposed in this paper uses particle filtering to estimate the SVR model parameters according to the whole measurement sequence up to the last observation instance. By treating the SVR training model as the observation equation of a particle filter, our method allows updating the SVR model parameters dynamically when a new observation comes. Because of the adaptability of the parameters to dynamic data pattern, the new PF–SVR method has superior prediction performance over that of standard SVR. Four application results show that PF–SVR is more robust than SVR to the decrease of the number of training data and the change of initial SVR parameter values. Also, even if there are trends in the test data different from those in the training data, the method can capture the changes, correct the SVR parameters and obtain good predictions.

ناشر
Database: Elsevier - ScienceDirect (ساینس دایرکت)
Journal: Reliability Engineering & System Safety - Volume 119, November 2013, Pages 109–116
نویسندگان
, , , ,