基于深度置信网络的松动部件质量估计方法研究

为解决以往核电站冷却系统松动部件质量估计方法存在的精度不高的问题,提出了一种基于深度置信网络的松动部件质量估计方法。基于平板模型上的不同质量钢球跌落实验,利用冲击信号的自功率谱与对应钢球质量来训练深度置信网络模型,进一步对跌落钢球质量进行了分类预测,并与支撑向量机和神经网络模型预测方法进行了比较。结果表明:深度置信网络方法能对跌落钢球质量进行较好的分类预测,分类平均正确率达到94%以上,预测结果好于支撑向量机(87.57%)和神经网络(91.64%),具有较高的跌落钢球质量预测精度。

Method of mass estimation of loose parts based on deep belief network

In order to solve the problem of low accuracy of the method for estimating the mass of loose parts in nuclear power plant cooling system,a method for estimating the loose parts mass based on deep belief network is proposed.Based on the different mass steel balls drop experiment on a plate model,the deep belief network model is trained by using the autopower spectrum of the impact signal and mass of the corresponding steel ball,and then the mass of the falling steel balls is further classified and predicted.This mass prediction method is compared with the support vector machine and neural network methods as well.Results show that the deep belief network method can better classify and predict the mass of the falling steel balls.The average accuracy of classification is over 94%.The prediction result is better than the 87.57%correct rate of the support vector machine and the 91.64%correct rate of the neural network.It can provide a feasible method for high-precision estimation of the mass of loose parts of nuclear power plant cooling primary circuit.