摘要
以252Cf中子源驱动噪声分析测量法为依据,利用中子脉冲信号自相关函数与被测核材料(252 U)质量的关系,设计了一种基于神经网络的核材料质量识别方法,探索借助时域特征进行质量识别的有效性。利用平稳小波变换抑制中子统计涨落对自相关函数带来的影响,利用分布式Elman神经网络对不同质量核材料的自相关函数样本进行训练和识别,并研究了有限样本前提下不同子网个数对最终识别结果所造成的影响。对4种核材料质量共计120组样本进行的实验,结果表明:在理想实验条件下,平稳小波变换抑制了统计涨落对信号自相关函数的影响;分布式Elman神经网络能够较好地识别自相关函数的特征,分辨不同质量的核材料,平均识别误差小于0.1。
According to the relationship between the autocorrelation function of neutron pulse signal and the mass of fissile material（252U）,this paper proposes an identification method for the mass of fissile material by means of artificial neural network and stationary wavelet transform.In order to suppress the ＂noise effect＂ of autocorrelation function due to statistical fluctuation of neutron signal,the wavelet approximation subband of the 2nd level is extracted after the autocorrelation function is decomposed,and the subband coefficients of different mass are reused as the input variables of distributed Elman neural network for training and recognizing.The impact of the number of subnetworks is also studied.The experimental results show that,under an ideal condition（4 kinds of mass and 120 sample functions）,the stationary wavelet transform overcomes the statistical fluctuations successfully and the distributed Elman neural network is able to distinguish different mass of fissile material with an average recognition error less than 0.1.
出处
《强激光与粒子束》
EI
CAS
CSCD
北大核心
2011年第10期2557-2559,共3页
High Power Laser and Particle Beams
基金
国家自然科学基金项目(61175005)
重庆大学研究生科技创新基金项目(CDJXS11120014)
关键词
252Cf中子源驱动噪声分析法
核材料识别
自相关函数
神经网络
平稳小波变换
252Cf neutron source driven noise analysis method
fissile material identification
autocorrelation function
neural network
stationary wavelet transform
