Predicting the spatial configuration of geological facies is a key step in the reservoir modeling process. The productivity of a reservoir depends not only on the facies proportions but also on the spatial patterns of the facies sequence. The recent developments in seismic to facies inversion techniques use \(1\mathrm{st}\)-order Markov models to improve the geological realism of the inferred facies profiles. However, the emergence of deep learning techniques such as recursive neural networks shows promising results in predictive modeling of event sequences as shown by the successful applications in complex modeling problems, such as natural language processing. In this work, a comparison between hidden Markov models and recursive neural networks is presented to highlight their advantages and disadvantages. The results are discussed according to the prior assumptions related to facies proportions and sequence patterns. Then, an innovative approach integrating recursive neural networks and the state-of-the-art seismic to facies inversion, known as the convolutional hidden Markov model, is proposed in order to predict geologically more realistic facies sequences based on seismic data. The proposed inversion technique is validated using synthetic seismic data in the context of a complex geological environment.

预测地质相的空间构造是储层建模过程中的关键步骤。储层的生产率不仅取决于相的比例，而且取决于相序的空间格局。地震至相转换技术的最新进展使用\（1 \ mathrm {st} \）阶马尔可夫模型，以改善推断相剖面的地质真实性。但是，诸如递归神经网络之类的深度学习技术的出现在事件序列的预测建模中显示出令人鼓舞的结果，如复杂建模问题（例如自然语言处理）中的成功应用所示。在这项工作中，提出了隐马尔可夫模型与递归神经网络之间的比较，以突出它们的优缺点。根据有关相比例和序列模式的先前假设讨论了结果。然后，采用一种创新的方法，将递归神经网络和最新的地震技术集成到相变中，称为卷积隐马尔可夫模型，为了预测基于地震数据的地质上更现实的相序，提出了一种方法。在复杂的地质环境中，使用合成地震数据对提出的反演技术进行了验证。