Feasibility study on transient identification in nuclear power plants using support vector machines

Support vector machines (SVMs), a relatively new paradigm in statistical learning theory, are studied for their potential to recognize transient behavior of detector signals corresponding to various accident events at nuclear power plants (NPPs). Transient classification is a major task for any computer-aided system for recognition of various malfunctions. The ability to identify the state of operation or events occurring at an NPP is crucial so that personnel can select adequate response actions. The Modular Accident Analysis Program, version 4 (MAAP4) is a program that can be used to model various normal and abnormal events in an NPP. This study uses MAAP signals describing various loss-of-coolant accidents in boiling water reactors. The simulated sensor readings corresponding to these events have been used to train and test SVM classifiers. SVM calculations have demonstrated that they can produce classifiers with good generalization ability for our data. This in, turn indicates that SVMs show promise as classifiers for the learning problem of identifying transients.