Support Vector Machines for Event Classification and Regression Analysis

The studies for developing the optimal machine learning algorithm such as artificial neural networks (ANNs), Bayesian inference, fuzzy inference, and support vector machines (SVMs) have been carried out. Among them, especially, the studies on application of SVMs to classification and regression problems is described in this paper. The current embodiment of SVMs was proposed by C. Cortes and V. Vapnik in 1995 and it is an algorithm with a neural network structure based on statistical learning theory. These SVMs have been generally used for event classification and identification. Using the kernel function, SVMs can effectively perform the nonlinear classification and regression analysis. There are studies using these SVMs in instrumentation and control field of nuclear power plants (NPPs). First of all, 7 transients of NPPs were classified and identified. Furthermore, golden time for accident recovery, power peaking factor (PPF), departure from nuclear boiling ratio (DNBR), residual stress of welding metal, and loss of coolant accident (LOCA) break size were estimated using SVM models. To be specific, the SVM method was used for a classification problem such as identification of the transients of NPPs and regression problems such as estimation of golden time for accident recovery, PPF, DNBR, and cutter wear, which shows good performance and the applicability.