一种基于CNN-LSTM神经网络的泥质烃源岩TOC预测模型

A TOC Prediction Model for Argillaceous Source Rocks Based on CNN-LSTM Neural Network

目前岩石中原始有机质的丰度可以使用测定岩石中残留的总有机碳含量(TOC)表示。但在实际应用时,受取心样品和实验分析成本的限制,单井烃源岩TOC的测定有限;同时受构造和沉积环境的控制,有机质的富集在纵向上变化也是比较大的。烃源岩富含有机质的煤系地层岩性差异大,测井响应特征受其影响变化大,因此该文避免岩性对测井曲线的影响,并且通过总结前人的研究经验,从神经网络出发,提出了一种利用测井曲线基于CNN-LSTM神经网络的泥质烃源岩TOC预测模型,经过实际应用,结果表明该方法有较好的准确性,可以得到较好的预测效果。

At present,the abundance of pristine organic matter present in the rock can be expressed by measuring the total organic carbon content(TOC)remaining in the rock.However,due to the limitation of coring samples and the cost of experimental analysis,TOC measurement of single well source rock is limited in practical application.At the same time,under the control of tectonic and sedimentary environment,the accumulation of organic matter in the longitudinal change is relatively large.The lithology of coal measures with rich organic matter in hydrocarbon source rocks varies greatly,and the logging response characteristics are greatly affected by it.Therefore,this paper avoids the impact of lithology on logging curves.Based on previous research experience and neural network,this paper proposes a TOC prediction model for argillaceous hydrocarbon source rocks using logging curves based on CNN-LSTM neural network.Through practical application,the results show that this method has good accuracy,good prediction results can be obtained.