Transient Identification in Nuclear Power Plants by PCA Based Neural Networks

Abstract

Nuclear power plants are complex engineering systems that are maintained and operated by human operators. When a transient occurs in an NPP, it is important for the operator to quickly identify and initiate corrective actions in order to mitigate the consequences, if any, and also to maintain the safety of the plant. As simultaneous generation of many alarms and annunciations in the control room may alert the operator on the onset of the transient it is quite possible that operator get confused to quickly act and identify the correct transient. By continuously analysing sensor data and detecting early warning signs, artificial intelligence models can provide timely alerts and enable operator to initiate proactive actions to mitigate the consequences of the transients. This paper presents a machine learning based transient identification model which assists the operator to identify the transient and initiate corrective action well within the stipulated time. The artificial neural networks based transient identification model has been developed for detecting loss of coolant accidents and main steam line break events in standard 220MWe Indian nuclear power plants. Several important input signals have been identified from the safety analysis reports and transient data has been generated from thermal–hydraulic simulation codes for training neural networks. The major outputs of neural network model are the size of the break, location of the break, etc. As the dimension of the input signals is significantly large, principal component analysis technique has also been employed to improve the performance of neural network model. The performance of neural network model with and without the application of principal component analysis is discussed in this paper.