遗传算法优化的BP神经网络在EDXRF中对钛铁元素含量的预测

在能量色散X荧光分析(EDXRF)技术中,受均匀效应、颗粒效应和基体效应等的干扰,定量分析精度受到影响。本文针对这一问题提出了遗传算法(GA)优化BP神经网络(GA-BP)的混合算法,该算法无需考虑元素浓度和射线强度之间的复杂关系。遗传算法优化BP神经网络的目的是为了获得更好的网络初始权值和阈值,其基本思想是:将初始化的BP神经网络均方根误差的倒数编码为遗传算法中个体的适应度;初始的权值和阈值用遗传算法中的个体代替,然后通过选择、交叉和变异操作挑选出最优个体,最后通过解码用最优的权值和阈值创建一个新的BP网络模型。攀枝花矿区5类矿样中钛和铁含量的整体预测和分类预测实验表明,分类预测效果远好于整体预测。预测值与化学分析值比较结果表明,其中76.7%的样品相对误差小于2%,表明了该方法在元素间基体效应校正上的有效性。     

Study on the characteristic points of boiling curve by using wavelet analysis and genetic neural network

Local singularity of a signal includes a lot of important information. Wavelet transform can overcome the shortages of Fourier analysis, i.e., the weak localization in the local time- and frequency-domains. It has the capacity to detect the characteristic points of boiling curves. Based on the wavelet analysis theory of signal singularity detection, Critical Heat Flux (CHF) and Minimum Film Boiling Starting Point (q_min) of boiling curves can be detected by using the wavelet modulus maxima detection. Moreover, a genetic neural network (GNN) model for predicting CHF is set up in this paper. The database used in the analysis is from the 1960s, including 2365 data points which cover a range of pressure (P), from 100 to 1000 kPa, mass flow rate (G) from 40 to 500 kg m⁻² s⁻¹, inlet sub-cooling (ΔT_sub) from 0 to 35 K, wall superheat (ΔT_sat) from 10 to 500 K and heat flux (Q) from 20 to 8000 kW m⁻². GNN mode has some advantages of its global optimal searching, quick convergence speed and solving non-linear problem. The methods of establishing the model and training of GNN are discussed particularly. The characteristic point predictions of boiling curve are investigated in detail by GNN. The results predicted by GNN have a good agreement with experimental data. At last, the main parametric trends of the CHF are analyzed by applying GNN. Simulation and analysis results show that the network model can effectively predict CHF.     

BP神经网络在DNBR计算中的应用

在压水堆事故分析中,通常采用系统分析程序、热流密度计算程序和子通道分析程序分步计算堆芯偏离泡核沸腾比(Departure from Nucleate Boiling Ratio,DNBR)。利用该方法计算的堆芯DNBR结果准确性较好,但是计算过程繁琐、费时。对于系统分析程序自带的堆芯DNBR简化计算模型,由于其根据堆芯限制线偏微分近似得到,适用范围较窄,准确性也难以保证。利用神经网络中的误差反向传播(Back Propagation,BP)算法,基于堆芯核功率、入口温度、流量和压力4个变量对应的一系列DNBR值,选取部分数据进行训练并建立模型,以达到快速和准确地预测堆芯DNBR的目的。根据误差分析,建立的计算模型具有较好的准确性,而且在部分失流事故和汽机停机事故下可较好地预测堆芯DNBR。     

EDXRF analysis of beach rock samples of Andaman Island

B Elemental contents of beach rock samples were analyzed using EDXRF. The samples were collected from three locations of Andaman Island. The Al, Ca, K, Fe, Ti, Si, V, Co, Cu, Ba, Zn, Ph, Cd and Mn contents were determined. The geochemical behavior of elements in the region is discussed. The elemental contents of beach rock samples from Andaman Island are much below the values of both earth crust and that of Tamilnadu region. However,content of the biogenic element Ca was the highest of all dements. This is due to the typical beach rock formation.     

LRAD测量管道内α污染活度的BP网络模拟

针对风速、流量、探源距、测试管管长和管径等因素对核设施退役中管道内α污染测量造成的非线性影响,采用控制变量法开展长距离α测量(Long range alpha detector,LRAD)模拟装置下的多参数影响实验,初步分析了各种因素对系统测量值的影响特征,建立了以影响因素和测量值为输入、放射源活度为输出的BP神经网络模型,分别对948和100组数据进行了模型建立和实例检验,结果说明,预测平均相对误差为3.4218×10–4,实例平均相对误差为2.217×10–2。应用BP网络模型模拟LRAD装置下的α活度是可行且有效的。     

基于MATLAB的γ谱人工神经网络分析方法

用人工神经网络分析γ能谱具有充分利用全谱信息，分析结果精度高的优点。在MATLAB平台上实现了该分析方法，具有编程简单、可靠、便于推广的优点。     

Neural network-based matrix effect correction in EDXRF analysis

In this paper we discuss neural network-based matrix effect correction in energy dispersive X-ray fluorescence (EDXRF) analysis, with detailed algorithm to classify the samples. The method can correct the matrix effect effectively through classifying the samples automatically, and influence of X-ray absorption and enhancement by major elements of the samples is reduced. Experiments for the complex matrix effect correction in EDXRF analysis of samples in Pangang showed improved accuracy of the elemental analysis result.     

BP神经网络算法预测多组分材料中子屏蔽效果方法研究

针对多组分中子屏蔽材料优化设计中蒙特卡罗模拟计算时间长而对算法效率的制约,讨论了利用BP神经网络算法快速预测材料中子屏蔽效果的方法。以复合材料300种随机质量组分和其对应的蒙特卡罗计算的剂量值组成训练样本,建立了典型的3层BP神经网络模型,其剂量预测值与样本值的绝对偏差在±2以内。对训练样本之外的验证样本,绝对偏差扩大到-6.4～5.2之间。偏差分布统计显示70%以上样本的相对偏差绝对值在2%以内,定性判断该神经网络模型的计算精度和泛化能力满足优化算法使用。使用交叉验证法对网络进行二次训练,可提高训练样本的计算精度,但扩大了验证样本的计算偏差,表明神经网络建立中还需要考虑样本的拟合程度和泛化能力的平衡。     

粒子群优化算法在神经网络识别γ能谱中的应用

在神经网络识别γ能谱的应用中,针对BP算法极易陷入局部极小、收敛速度慢的缺点,根据粒子群优化算法具有全局寻优的特点,本文将PSO与BP算法结合起来形成一种训练神经网络的新算法--混合PSO-BP算法.将该算法应用到γ能谱识别中,克服了BP算法极易陷入局部极小的缺点,并且训练好的网络具有很好的泛化能力,识别正确率为100%.实例表明,混合PSO-BP算法用于γ能谱识别是非常理想的、有效的.     

遗传神经网络在蒸汽发生器故障诊断中的应用

针对传统的BP神经网络学习算法易陷入局部极小以及收敛速度慢等问题,本文在神经网络中融合遗传算法,并将其应用到蒸汽发生器(SG)故障诊断中。结果证明,该算法能有效地解决网络训练中的收敛问题。     

Crack depth estimation of non-magnetic material by convolutional neural network analysis of eddy current testing signal

ABSTRACT

When a heat transfer tube of the steam generator of a pressurized water reactor fails, the primary cooling water leaks quickly into the secondary system. Moreover, if this leakage is large, the nuclear reactor emergency core cooling system (ECCS) may be activated. In Japan, to prevent such situation to take place, periodic inspections are performed in order to check whether heat transfer tubes are cracked. Eddy Current Testing (ECT) is a type of non-destructive inspection method used to detect cracks in a conductive material. ECT can estimate the shape of a crack by inverse problem analysis, but it is computationally expensive. Therefore, in this study, we aimed to develop a method to estimate crack depth by Convolutional Neural Network (CNN). The method was shown to be less computationally expensive during estimation and was robust against lift-off fluctuation during measurements.     

BP网络和OLAM网络在X荧光谱分析中的应用比较

用计算机软件实现了两种神经网络：BP和OLAM，并将两种网络模型用于水泥生料样品的X荧光谱分析，其中OLAM得到了较好的效果，从理论和实际的角度对两种算法的优劣和适用范围进行了讨论。     

BP神经网络预测EDXRF中铁、钛元素含量

为了更好地定量分析矿石样品中铁、钛元素的含量,应用EDXRF分析技术建立了一个基于BP神经网络的预测模型。将矿石样品在EDXRF光谱仪中测得的荧光强度计数作为BP神经网络模型的输入变量,对该模型进行训练和检测。实验结果表明:该BP神经网络预测模型能获得较精确的结果,预测铁含量结果的相对误差不大于2.4%;预测钛含量结果的最大相对误差不大于2.3%;可用于地质矿石样品元素含量的分析预测。     

Pollution assessment in the Trancão river basin (Portugal) by PIXE, EDXRF and isotopic analysis

This paper reports on the inorganic contamination assessment of a river basin and local water resources in order to establish quality standards. PIXE was applied to the elemental determination of the freeze-dried water dry residue and EDXRF was used for the evaluation of elemental content of sediments. To infer the water provenance the electrical conductivity was measured as well as the isotopic composition of surface waters using the ¹⁸O/¹⁶O ratio. The combined isotopic and elemental composition information enables to establish dominant contamination contributions from the several tributaries. Moreover, the variability observed for certain parameters, associates them with specific basin regions as for instance, dry residue mass, conductivity and Br, or S, Cl and As or Cr, Ni, Zn and Pb which permit to establish both pollution characterisation and their origin (agriculture, industrial, etc.). At certain locations, enhanced concentrations of elements as Cr, Cu, Zn and Pb are observed both in sediments and in the surface water. The elemental particular associations also permit to characterise pollution sources.     

鉴别暗物质粒子WIMP的人工神经网络方法

首次把人工神经网络方法用于冷暗物质的寻找实验。克服在低能区间中单变量方法的不适应性。对标定数据的分析结果表明，神经网络的分类效率达到75％左右，且训练表现和泛化表现完全一致，这个结果是单变量方法所无法企及的。因而可以相信，人工神经网络方法用于暗物质粒子鉴别将是一个很有前景的课题。     

Quantitative analysis modeling for the ChemCam spectral data based on laser-induced breakdown spectroscopy using convolutional neural network

Laser-induced breakdown spectroscopy (LIBS) has been applied to many fields for the quantitative analysis of diverse materials. Improving the prediction accuracy of LIBS regression models is still of great significance for the Mars exploration in the near future. In this study, we explored the quantitative analysis of LIBS for the one-dimensional ChemCam (an instrument containing a LIBS spectrometer and a Remote Micro-Imager) spectral data whose spectra are produced by the ChemCam team using LIBS under the Mars-like atmospheric conditions. We constructed a convolutional neural network (CNN) regression model with unified parameters for all oxides, which is efficient and concise. CNN that has the excellent capability of feature extraction can effectively overcome the chemical matrix effects that impede the prediction accuracy of regression models. Firstly, we explored the effects of four activation functions on the performance of the CNN model. The results show that the CNN model with the hyperbolic tangent (tanh) function outperforms the CNN models with the other activation functions (the rectified linear unit function, the linear function and the Sigmoid function). Secondly, we compared the performance among the CNN models using different optimization methods. The CNN model with the stochastic gradient descent optimization and the initial learning rate=0.0005 achieves satisfactory performance compared to the other CNN models. Finally, we compared the performance of the CNN model, the model based on support vector regression (SVR) and the model based on partial least square regression (PLSR). The results exhibit the CNN model is superior to the SVR model and the PLSR model for all oxides. Based on the above analysis, we conclude the CNN regression model can effectively improve the prediction accuracy of LIBS.     

一种基于BP神经网络的γ能谱识别方法

从军控现场核查过程中敏感信息保护的目的出发,对利用BP神经网络技术通过γ能谱识别核部件类型的方法进行了初步研究.实验时识别正确率达到98%以上,说明利用神经网络技术识别核部件是一种切实可行的方法.     

人工神经网络和最小二乘回归在XRF定量分析中的应用比较

采用人工神经网络的误差反向传输算法,定量分析了Ｐｔ、Ｐｄ二元合金的荧光Ｘ射线能谱,并与最小二乘回归法进行了比较。     

人工神经网络在核供热堆故障诊断中的应用研究

运用人工神经网络技术对２００ＭＷ核供热堆的故障诊断系统进行了研究,并用事故工况下反应堆参数的实际值和趋势变化值分别对两个ＢＰ网络进行训练和检验,两个网络诊断结果的综合得出最终诊断结果。经检验证明,将两个网络结合的综合系统与单网络系统相比,可提高诊断的准确性和适应性。     

基于多准则神经网络优化的Compton背散射成像

详细阐述了基于散射能量谱的行扫描Compton背散射成像原理，将成像方程归结为一个大型非线性方程组，然后用基于多目标优化的神经网络来求出重建问题的解。数据模拟结果表明，所构造的神经网络对Compton背散射多准则成像是可行的，适于处理不适定重建问题。