混合神经网络的核电站故障程度评估方法
Hybrid neural network for evaluating the fault degree of nuclear power plants
周桂 1王航 1彭敏俊1
作者信息
- 1. 哈尔滨工程大学 核安全与仿真技术重点学科实验室,黑龙江 哈尔滨 150001
- 折叠
摘要
为了准确处理大量复杂数据,避免由于压力上升导致的操纵员判断失误,本文提出了一种基于改进的粒子群算法和留一交叉验证的混合神经网络超参数优化方法,辅助操纵员评估核电站故障程度.该方法通过改进的粒子群算法优化超参数组合,利用留一交叉验证评估深度学习模型泛化性能,最终构建高精度故障程度评估模型.本文以核电站失水事故为对象,对所提出方法进行测试验证.结果表明:本文提出的混合神经网络超参数优化方法能够搜索最优超参数组合,构建绝对精度为 97%的神经网络模型,能有效评估核电站故障程度,辅助操纵员维修决策.
Abstract
To accurately process a large amount of complex data and avoid operator judgment errors caused by in-creased operator pressure,this paper proposed a hybrid neural network hyperparameter optimization method based on an improved particle swarm optimization algorithm and leave-one-out cross validation to assist operators in evalu-ating the degree of failure in nuclear power plants.This method optimized hyperparameter combinations through an improved particle swarm optimization algorithm,evaluated the generalization performance of deep learning models through leave-one-out cross validation,and ultimately constructed a high-precision failure severity evaluation mod-el.The proposed method was tested and evaluated by considering the loss of coolant accident condition.The results show that the proposed method for evaluating the degree of nuclear power plant failure based on hybrid neural net-work hyperparameter optimization can search for the optimal combination of hyperparameters,achieving an absolute accuracy of 97%in neural network modeling and effectively evaluating the failure degree of the nuclear power plant to assist the operator in maintenance decisions.
关键词
核电站/故障程度评估/超参数优化/混合神经网络/粒子群算法/留一交叉验证/破口事故/操纵员决策Key words
nuclear power plant/fault degree evaluation/hyperparameter optimization/hybrid neural network/par-ticle swarm optimization/leave-one-out cross validation/loss of coolant accident/operator decision引用本文复制引用
基金项目
工业和信息化部核能开发项目(KY11500200118)
出版年
2023
国家科技期刊平台
NSTL
维普
万方数据