针对核磁共振永磁体的磁场传统匀场方法难度大，经验依赖程度高的情况，提出了人工神经网络应用于磁场匀场的手段。通过基于Matlab的神经网络设计，建立了一个匀场的BP网络模型。由于在磁体的不同的均匀度条件下匀场的方法不同，所以该网络由两个子网络组成----粗调网络和精调网络，并分别对其进行训练。该模型建立之后，可以快速、准确地匀场，使无源匀场不再盲目得进行多次尝试，一次就能使判断准确。通过人工神经网络方法匀场的磁体，均匀度高，满足了核磁共振岩心分析仪的匀场环境。

: Homogeneous magnetic field is necessary in NMR magnet. In general shimming magnetic field is by experience. So it’s difficult to shim. It depends on experiences frequently. Artificial Neural Network based on the ANN toolbox of Matlab is put forward to apply in the process. Error backward propagating network is created. The method is varied from different uniformity of magnetic field. The neural network consists of two subnets—coarse regulating network and fine regulating network which are trained respectively. After the model is established, shimming magnet becomes rapidly and accurately. The ferromagnetic shim needn’t be tried many times and aimlessly. It is judged only once. Magnet shimmed by ANN has a nice uniformity and can meet the condition of NMR core Analyzer well.