کد مقاله کد نشریه سال انتشار مقاله انگلیسی نسخه تمام متن
6955272 1451855 2016 14 صفحه PDF دانلود رایگان
عنوان انگلیسی مقاله ISI
An adaptive online learning approach for Support Vector Regression: Online-SVR-FID
موضوعات مرتبط
مهندسی و علوم پایه مهندسی کامپیوتر پردازش سیگنال
پیش نمایش صفحه اول مقاله
An adaptive online learning approach for Support Vector Regression: Online-SVR-FID
چکیده انگلیسی
Support Vector Regression (SVR) is a popular supervised data-driven approach for building empirical models from available data. Like all data-driven methods, under non-stationary environmental and operational conditions it needs to be provided with adaptive learning capabilities, which might become computationally burdensome with large datasets cumulating dynamically. In this paper, a cost-efficient online adaptive learning approach is proposed for SVR by combining Feature Vector Selection (FVS) and Incremental and Decremental Learning. The proposed approach adaptively modifies the model only when different pattern drifts are detected according to proposed criteria. Two tolerance parameters are introduced in the approach to control the computational complexity, reduce the influence of the intrinsic noise in the data and avoid the overfitting problem of SVR. Comparisons of the prediction results is made with other online learning approaches e.g. NORMA, SOGA, KRLS, Incremental Learning, on several artificial datasets and a real case study concerning time series prediction based on data recorded on a component of a nuclear power generation system. The performance indicators MSE and MARE computed on the test dataset demonstrate the efficiency of the proposed online learning method.
ناشر
Database: Elsevier - ScienceDirect (ساینس دایرکت)
Journal: Mechanical Systems and Signal Processing - Volumes 76–77, August 2016, Pages 796-809
نویسندگان
, ,