Quantitative Interpretation Research Of Gamma Spectrometry Logging In Uranium Exploration Based On Machine Learning Algorithm

Uranium ore is an important nuclear fuel resource,and it is of great practical importance to carry out γ-energy spectrometry logging for uranium ore.This paper mainly focuses on the quantitative interpretation of the radioactive isotopes in gamma spectrum obtained from uranium mine logging.In order to solve the problem of low statistical performance spectral analysis to improve the logging speed and to eliminate the interference of spectrum drift on the quantitative interpretation of energy spectrum,this paper proposes a research method based on multiple machine learning algorithms for gamma spectrum quantization interpretation.The main research contents and achievements are as follows:(1)BP neural network,GRNN,and SVM machine learning algorithm models were constructed to solve the problem of U,Th,and K quantitative interpretation in gamma-ray logging.Based on the measured data of the uranium mine standard model well in the Aerial Survey Remote Sensing Center of Nuclear Industry,the performance of three machine algorithms was analyzed comprehensively.Results showed that BP neural network had relatively good quantitative interpretation results for U,Th,and K,with a mean squared error of 0.2333 and an average absolute error of 0.1699.The interpretation results of low-grade and medium-high-grade uranium ores differed,and the quantitative accuracy was 86.986% and 93.478%,respectively.(2)A method for analyzing high logging speed and spectrum drift in gamma spectrum based on machine learning algorithms was proposed.By comparing the machine learning algorithm analysis results under different logging speeds and spectrum drift conditions,it was found that as the spectral statistical effect and degree of spectrum drift gradually improved,the quantitative error of the model changed less,indicating that the gamma spectrometry logging in uranium exploration quantitative interpretation based on machine learning algorithms can well overcome the effects of logging speed and spectrum drift,which is conducive to improving the traditional logging speed to 6 m/min.(3)An optimized gamma spectrum quantitative interpretation method was established.For the spectrum with a logging speed of 6 m/min,a U quantitative interpretation accuracy of 99.176% was achieved in the case of medium-high-grade U content and a relative quantitative error of less than 10%,which was 5.7% higher than that of BP algorithm.Analyzing low statistical gamma spectrum based on machine learning algorithms can help improve logging speed.In the case of higher logging speed,this study provides a new approach for the selection of gamma spectrometry logging in uranium exploration quantitative interpretation methods.It also provides a reference for mineral analysis and uranium resource evaluation from another perspective.