Prediction of components degradation using support vector regression with optimized parameters

Abstract To solve the problem that traditional data-driven methods cannot acquire high accuracy with few monitoring data in nuclear power component degradation prediction, this article proposed the support vector regression (SVR) algorithm and provided a hybrid parameters optimization strategy for SVR using grid search and cross validation. This article analyzed the performance of SVR algorithm in three different cases: function fitting, multivariate regression, and prediction of nuclear power plant pipeline corrosion. Results show that the performance of proposed SVR algorithm is better than classical BP neutral network by comparing prediction mean square error and squared correlation coefficient of two methods. The advantage of SVR algorithm in small-sample learning is also demonstrated and it turns out that the SVR algorithm is an effective approach for modelling component degradation with rare inspection data.