Optimal lunar internal structure model obtained by a neural network method [基于神经网络方法获得最优化月球内部结构模型]

作者：Liao B.; Xu J.; Chen X.; Sun H.; Zhou J.

来源：Acta Geophysica Sinica, 2022, 65(3): 939-951.

DOI：10.6038/cjg2022O0237

• Bayesian inverse problem
• Lunar internal structure
• Lunar low-velocity zone
• Mixture Density Neural Network (MDN)
• Spectral element method

Because of limited observation techniques, current understanding of the lunar internal structure is so uncertain that no lunar internal structure model has been widely recognized and adopted until nowadays. In addition, little attention had been paid to the effects of different observations accuracies on the structure model. Using the mixture density neural network (MDN) method, the posterior probability density distributions of the lunar models are obtained in this study. Based on the density distribution, the average lunar internal structure model (Mean model), the maximum posterior probability model (MAP model) and the lunar model range satisfying the 1-σ (1-σ Model) are computed, and the MAP model is the optimal lunar internal structure model given in this paper. In addition, our results show that the S-wave velocity in the low-velocity zone of the moon is lower than that of the lunar mantle, which supports the opinion of the existence of a low-velocity zone at the bottom of the lunar mantle. Numerical results of the influences of different observation accuracies on lunar model show that the Love number k2 has a lower boundary with a value of about 0.0220. In addition, the accuracy of k2 has a significant influence on the lunar model comparing with influences of the average density and the average moment of inertia.

• 单位
  大地测量与地球动力学国家重点实验室; 中国科学院大学