A Support Vector Machine Model for Detection of Transients in Nuclear Reactor

For safe and efficient operation of nuclear reactors, important parameters are measured by a number of sensors enabling the operator to monitor and diagnose operational anomalies. However, during a transient, many signals and alarms occur simultaneously in control room and operators face pressure to assess trends and alarms across a range of parameters. As a result, the operators are challenged to immediately identify evolving anomalous events and identify the precise transient. Operators may thus take improper decisions and actions that may result in unwanted events. A highly helpful tool will be machine learning models that provide transient detection based on reactor signals. In this study, a Support Vector Machine (SVM) model for the detection of transients in a pool type research reactor is proposed. SVM is a supervised machine learning algorithm which segregates an n-dimensional space into classes by using a hyperplane. For data-driven modeling, simulated datasets representing major transients anticipated in a pool type research reactor, including Loss of Coolant and Loss of Flow, are used. Important plant parameters like power, coolant flow & reactor pool level, measured and acquired in the plant's Computerized Operator Information System are used for modeling. The SVM model is trained and tested to distinguish the transients with a high degree of classification accuracy. The proposed SVM model can detect hypothesized transients of a general pool reactor. This will assist the operator identify anomalous plant states and take corrective action.