核电站失水事故的智能预警及仿真方法研究

Research on Intelligent Early Warning and Simulation Methods for Loss of Water Accidents in Nuclear Power Plants

使用基于深度学习的卷积神经网络(Convolutional Neural Network,CNN)和卷积长短期记忆(Convolutional Long-Short Term Memory,ConvLSTM)模型进行核电站失水事故(Loss of Coolant Accident,LOCA)的预警及仿真综合模型的构建。利用CNN的特征提取能力有效识别不同破口尺寸的相关特征,并对工况发展进行预测分类,判定事故发生可能性并给出事故预警。利用预警阶段生成的事故种类判定,使用ConvLSTM在给定时长中计算LOCA工况中的关键系统参数变化情况,实现基于深度学习的LOCA工况仿真。多种方式的实验验证了该模型较好的功能性和适应性,事故类型判定正确率达到88%,仿真工况与原始值的误差保持在10-5量级。利用深度学习模型在特征提取和数值拟合方面的能力,将来还可以对核电站的工况仿真与故障分析进行进一步的智能化改进。

This article uses deep learning-based Convolutional Neural Network(Convolutional Neural Network,CNN)and Convolutional Long-Short Term Memory(Convolutional Long-Short Term Memory,ConvLSTM)models for nuclear power plant loss of water accident(Loss of Coolant Accident,LOCA)Construction of a comprehensive model of early warning and simulation.Use the feature extraction capability of CNN to effectively identify the relevant features of different breach sizes,predict and classify the development of working conditions,determine the possibility of accidents and give accident warnings.Using the type of accident generated in the early warning stage,ConvLSTM is used to calculate the change of key system parameters in the LOCA working condition for a given period of time to realize the LOCA working condition simulation based on deep learning.Various experiments have verified the good functionality and adaptability of the model.The accuracy of the accident type determination is 88%,and the error between the simulated operating conditions and the original value is maintained at the order of 10-5.Utilizing the capabilities of deep learning models in feature extraction and numerical fitting,further intelligent improvements can be made to the simulation and fault analysis of nuclear power plants in the future.