PCA-Based Neural Network Model for Identification of Loss of Coolant Accidents in Nuclear Power Plants

Abstract

Nuclear power plants (NPPs) are extremely complex systems that are operated and monitored by human operators. Maximum care is exercised to keep the likelihood of potential risks to a very low value. However, in the event of an unlikely abnormal occurrence, the operator has to take necessary actions relatively fast, which involves complex judgments, making trade-offs between partly incompatible demands, and requires expertise to take proper decision. Over the years, several intelligent systems have evolved to assist the operator for decision-making; however they are highly computationally intensive and may not be suitable for real-time online monitoring or may require large amounts of data. In this paper, an efficient artificial neural network (ANN) model has been developed based on principal component analysis (PCA) for identification of large break loss-of-coolant accident (LOCA) in NPPs. A large database of reactor process parameters is generated through various thermal hydraulic codes, and PCA was performed for 32 break scenarios of LOCA in inlet and outlet reactor headers with and without the availability of emergency core cooling system (ECCS). The PCA was used to optimize the inputs of ANNs. The results of comparison between the classical and PCA-based ANN have been presented in this paper. The simplified ANN model based on PCA is relatively in good agreement with the classical ANN model. It can be said that the PCA-based ANN gives a great computational advantage, due to an important factor when the input parameter dimension is substantially optimized and is usually a case in NPPs. However, there is a scope of improvement in the PCA-based ANN in terms of reduction of error, and this could be achieved by incorporating more of variance during dimension reduction by PCA and also applying different architectures of ANN.