基于深度学习的煤质组分测井解释方法

Logging Interpretation of Coal Components Based on Deep Learning

为了解决目前煤质组分的测井定量解释中存在的各种问题,引入了深度学习的方法,利用具备实验室分析结果的测井数据建立煤质组分含量的深度信念网络,通过实验,探索网络参数对计算结果的影响。结果表明,在深度信念网络中,限制波尔兹曼机层数的选取需要兼顾计算精度和速度,隐含层神经元个数需要兼顾计算精度和稳定性,激活函数以Re LU函数为最佳。

In order to solve the problems of the logging quantitative interpretation in coal components, introduces the deep learning, and applies logging data with results of laboratory analysis to establish the deep belief network（DBN） that can be used to compute the coal components, then explore the effects of network parameters on the calculations by experiment. The results show that in DBN, the number of restricted Boltzmann machine should be determined by the calculations precision and speed; the neurons＇ number of hidden layers need to balance the accuracy and stability of calculations; and the best activation function is ReLU.