Liquid Scintillator N-γ Discrimination Based On Deep Learning And Temperature Correction For Attenuation Length Measurement Systems

Liquid scintillators are widely used in particle detector such as neutron detectors and neutrino detector,because they have high light yield and optical performance,strong compatibility with containers,can change shape with the shape of containers,and have the advantages of low toxicity,low cost,high flash point,and high environmental protection.As a new type of scintillator,liquid scintillators have very broad application prospects.Therefore,studying their optical and physical properties can promote their various applications.Liquid scintillator neutron detectors in the process of neutron detection,often accompanied by γ-ray background,in the work of n-γ discrimination,due to the increase in count rate,electronics instrument range limitations and other factors to produce piled-up pulses and overflowed pulses,the pulse structure of these events is substantially changed.In the face of the complex waveforms in the detection process of liquid scintillator neutron detectors,some of the traditional discrimination algorithms,such as charge comparison method,are unable to meet the requirements of accurate discrimination in terms of their adaptiveness and robustness,therefore,there is an urgent need for an algorithm with high adaptiveness and robustness for more effective discrimination.In this study,a Res Net neural network algorithm is proposed to address the above problems for accurate discrimination of n-γ pulse signals in complex circumstances.Firstly,a novel liquid scintillator(DIN)is used in this study to form a neutron detection platform to collect n-γ pulse signals,and the charge comparison method is used to make pulse class labels for the collected pulse signals.Since the recognition performance of the neural network in complex circumstances needs to be investigated,the possible piled-up events and overflowed events in complex circumstances are subsequently simulated to constitute a composite data set.Finally,the dataset was trained and validated using the proposed three different types of neural networks MLP,CNN and Res Net.The experimental results show that Res Net has the lowest number of error discrimination in the confusion matrix.In the F1 score,Res Net also has the largest F1 score.In the ROC curve and AUC,Res Net is still the best performing method.Res Net significantly enhances the adaptiveness and generalization ability required in this work.By cross-sectional comparison,the Res Net proposed in this paper is significantly improved in this work and is more suitable for n-γ discrimination in complex circumstances.During the construction of the Jiangmen Underground Neutrino Observatory(JUNO),the central detector will use nearly 20,000 tons of liquid scintillators,so it is crucial to carry out accurate measurement of the attenuation length of liquid scintillation solvents and finished products in all aspects of liquid scintillation development and production.The existing attenuation length measurement system works with a certain systematic error due to the mechanical accuracy of the optical instrument and the nature of the photomultiplier tube.After debugging and scaling,the measurement errors caused by problems such as spot displacement and PMT photocathode unevenness have been optimized under the existing mechanical accuracy.Therefore,in order to enhance the credibility of measurement results and reduce the systematic error,this study will reduce the effect of the photomultiplier tube on the attenuation length measurement system by studying its temperature effect.First,the temperature-light intensity amplitude relationship is evaluated by Pearson correlation coefficient,and a strong linear negative correlation between temperature and measured value can be obtained,and the effect of temperature on measured value is-12.3/°C.The linear correction of the measurement results can be performed by monitoring this relationship with the real-time room temperature.After the correction,29 rounds of measurements and 8 rounds of measurements were performed on different samples to verify that the mean difference was only 0.13 m in 29 rounds of measurements,and the standard deviation was reduced by 74.6% after the correction.In 8 rounds of measurements,the mean difference was 0.63 m,and the standard deviation was reduced by 82% after the correction.Finally,the comparison of the measurement verification before and after the correction concluded that the stability of the single-round measurement and the data reliability of the temperature-corrected attenuation length measurement system were greatly improved.