基于优化遗传神经网络方法对α能谱去噪的研究

Investigation of Alpha Spectroscopy Noise Subtraction Based on Optimum Genetic Nerve Network Method

为实现准确快速得到去除噪声后α射线谱,采用了一套可以让探测器处于不同真空条件下的探测设备。将实验研究获得的α能谱中能表征能谱信息的特征参数如峰值、不同高度的左右半宽度和左右边界,在遗传算法中经选择、交叉操作进行权值和阈值的优化后,作为输入层数据对BP神经网络进行训练。经过反复训练最终得到去噪后的输出结果。实验结果表明:对于α能谱的去噪,遗传算法优化后的BP神经网络比未被优化的效果要好。

In order to approach an accurate and immediate alpha spectrometry with noise subtraction,a set up with different vacuum conditions was used. Several parameters obtained from experimental alpha spectroscopy and indicating properties of spectroscopy, e. g. peak value, half-peak width and boundaries,were optimized in terms of weight and threshold by selecting and cross-operating in genetic algorithm prior to be trained in BP nerve network as input data. Experimental results show that the optimized BP nerve network is better than the non-optimized one in terms of alpha spectroscopy noise subtraction.