针对252Cf自发裂变中子源构成的核信息系统，以实际所测随机中子脉冲数据的自相关函数为研究对象，借助仿真实验， 利用Elman神经网络对不同质量核材料进行识别。在实测数据的基础上，通过叠加随机抖动，模拟产生了不同质量核材料的相关函数样本，并将其用于神经网络的训练与测试，实验结果表明，训练过的Elman神经网络能够较好地识别相关函数的特征，分辨不同质量的核材料，平均识别率达到85%，综合平均误差为0.04，且具有较高的鲁棒性。

Correct and effective identification of nuclear material is a key part for nuclear weapon’s verification. Based on the nuclear information system constructed with 252Cf spontaneous neutron source, the simulated autocorrelation functions are utilized for identification of different fissile mass with Elman neural network. The experimental results show the trained Elman neural network is able to distinguish the characteristics of autocorrelation function, identify different fissile mass, and the average identification rate reaches 85% and keep average synthesis error smaller than 0.04 with high robustness.