【摘要】 随着核电厂向数字化和智能化转型，利用神经网络对瞬态参数进行预测，辅助操作人员处理事故成为可能。针对基于梯度下降的BP神经网络在预测核电厂瞬态参数时可能陷入局部最优的问题，提出了一种结合萤火虫算法(Firefly Algorithm, FA)优化的BP神经网络(FA-BP神经网络)。使用PCTRAN仿真软件生成的数据，比较了FA-BP神经网络与传统BP网络在预测性能上的差异，并应用FA-BP神经网络进行故障诊断。研究结果表明，FA-BP神经网络在训练效率和预测精度方面均显著优于传统BP网络，并在故障诊断中展现出高准确率。实验表明FA-BP模型能够支持核电厂操作人员在事故中更有效地管理机组状态，增强核电安全性。更多还原

【Abstract】 As nuclear power plant transition towards digitalization and intelligence, utilizing neural networks to predict transient parameters to assist operators in handling incidents and accidents has become feasible. This study solves the problem of gradient descent-based BP neural networks potentially falling into local optima when predicting transient parameters of nuclear power plant, by proposing a BP neural network optimized with the firefly algorithm(FA-BP). Data generated by the PCTRAN simulation software were used to compare the predictive performance between the FA-BP neural network and traditional BP network, and FA-BP neural network was applied to fault diagnosis. The results indicate that the FA-BP neural network significantly outperforms the traditional BP network in both training efficiency and predictive accuracy and demonstrates high accuracy in fault diagnosis. The experiments show that the FA-BP model can more effectively support nuclear power plant operators in managing unit states during incidents and accidents, thereby enhancing nuclear power safety.更多还原