基于长短时记忆神经网络LSTM的煤体结构识别方法

Coal structure recognition method based on long short memory neural network LSTM

煤体结构的定量识别是煤层和煤层气等资源勘探开发中的关键性问题之一。本次研究以鄂尔多斯盆地A地区煤岩为研究对象,选取5条常规测井曲线井径CAL、自然伽马GR、声波时差AC、补偿中子CNL和密度DEN曲线,构建基于LSTM神经网络的煤体结构识别模型,实现了煤体结构的定量识别。研究表明,随着煤层被破坏程度的增大,构造煤具有DEN曲线和GR曲线值偏低,AC曲线、CNL曲线和CAL曲线值偏大的特征;LSTM网络模型参数是影响模型识别准确度的重要因素,通过多次试验得出网络训练的迭代次数、LSTM神经网络包含的隐藏单元数和数据最小分块尺寸参数的最优值分别为800、32和28;基于LSTM神经网络的煤体结构识别模型准确度达到85.5%,并且利用未参与模型构建的验证井进行可靠性的验证,认为该方法可有效识别煤体结构,基本满足实际生产的需求,对于相似地区煤体结构的识别也具有一定的借鉴意义。

The quantitative identification of coal structure is one of the key problems in the exploration and development of coal seam and coalbed methane resources.This study takes the coal and rock in area a of Ordos basin as the research object,selects 5 conventional logging curves,well diameter cal,natural gamma GR,acoustic time difference AC,compensated neutron CNL and density den curve,constructs the coal structure identification model based on LSTM neural network,and realizes the quantitative identification of coal structure.The results show that with the increase of the damage degree of coal seam,the values of Den curve and GR curve are low,and the values of AC curve,CNL curve and cal curve are high;LSTM network model parameters are important factors affecting the accuracy of model recognition.Through many experiments,it is concluded that the optimal values of the number of iterations of network training,the number of hidden units contained in LSTM neural network and the minimum block size of data are 800,32 and 28 respectively;the accuracy of the coal structure identification model based on LSTM neural network is 85.5%,and the reliability is verified by using the verification well that does not participate in the model construction.It is considered that this method can effectively identify the coal structure,basically meet the needs of actual production,and has a certain reference significance for the identification of coal structure in similar areas.