Optimization of BP neural network for fault parameter prediction in nuclear power plants utilizing the firefly algorithm

The digitalization and intelligence of nuclear power plants (NPPs) enable neural networks to predict transient parameters, aiding operators in emergency responses. This paper addresses the limitations of the back propagation (BP) neural network, which can lead to local optima and reduced accuracy in predicting transient parameters, by introducing the Firefly Algorithm (FA) to optimize the BP network, creating the FA–BP neural network. Using transient data from PCTRAN, a nuclear power plant accident simulation software, the study compares the predictive performance of the FA–BP network with conventional BP and LSTM neural networks during loss of coolant accident and steam generator tube rupture accidents. The results show that the FA–BP network reduces prediction errors and achieves similar accuracy with shorter prediction times compared to LSTM, due to its simpler structure. The FA's global optimization capabilities enhance the BP network's performance, making it more effective in predicting long-term post-accident parameters. The FA–BP neural network model offers a promising approach for improving the accuracy and speed of transient parameter predictions, contributing to enhanced safety and control in nuclear power plants during emergencies.