FFT-BP神经网络模型对车载γ能谱辐射剂量率的预测分析

摘要

In order to measure the radiation dose rate accurately with carborne γ spectrometer patrol system,proposed a modified back-propagation network(BP network)model basised on fast fourier transform background deduct method(FFT-BP network model).Using γ-ray energy spectrum analysis method to test the carborne γ-ray energe spectrum of Cs-137 of different spac-ings,adopting FFT method to deduct the background of spectrum data then get new spectrum data.The modified B-P network model is applied to qualitatively predict the radiation dose rate of unknow dose carborne γ spectrum,by comparing the predicted results with fitting results of 3 function models to verify the effect of FFT-BP network model.The results show the FFT deduct method can weaken the influence of the scattering background on γ spectrum and reduce spectrum background effectively.The correlation coefficients between characteristic peak area and net area getting from new spectrum are 0.99(p< 0.05),which shows a remarkable correlation.In the process of model fitting,FFT-BP network model shows strong ability of learning and generalization,the prediction of experimental results is ideal,relative error and accumulative error are below 0.6% and 9% re-spectively,it has better effect than mathematical methods and gamma spectra method and it also can reduce the error of radiation dose rate analized by γ-spectra analysis method,improve the work efficiency effectively.There fore,FFT-BP network model can apply to predictive analysis of γ-ray energy spectrum radiation dose,which provide a new and efficient method for carborne γ spectrometer patrol system to measure radiation dose.%为了实现车载γ谱仪巡测系统对辐射剂量率的准确测定,提出基于快速傅里叶变换(FFT)本底扣除法的改进型BP神经网络模型(FFT-BP神经网络模型).实验采用γ射线能谱分析法,对不同间距处的137 Cs放射源进行车载γ能谱测试,将得到的谱数据通过快速傅里叶变换(FFT)扣除本底,获得新的谱线数据.应用FFT-BP神经网络模型对未知剂量的车载γ谱线作辐射剂量率的定量预测,将预测结果同3个函数模型的拟合结果比较,验证FFT-BP神经网络模型的预测效果.结果表明,FFT扣除法能较好的削弱散射本底对γ谱线的影响,能有效的降低谱线本底.通过新谱线获得的特征峰面积和净谱线面积与辐射剂量率的相关系数均为0.99(p<0.05),相关性显著.模型拟合分析过程中,FFT-BP神经网络模型表现出较强的学习泛化能力,预测较理想,相对误差和累计误差分别低于0.6% 和9%,效果明显优于数学模型和γ能谱全能峰法,可显著降低γ能谱分析辐射剂量率的误差,且能有效提升工作效率.因此,FFT-BP神经网络模型适用于γ能谱辐射剂量的预测分析,为车载γ谱仪巡测系统测量辐射剂量提供了一种新型有效的分析方法.