Lithology identification method based on integrated K-means clustering and meta-object representation

Aiming at the problem that the basic identification unit constructed by the existing lithology identification method cannot fully utilize the context information provided by the continuity of logging curves and the potential information in each layer segment, a lithology identification method based on integrated K-means clustering and meta-object representation is proposed. The K-means clustering method is used to automatically layer multiple conventional logging curves of the target reservoir, and the obtained layer is used to construct a meta-object for lithology identification. Specifically, in order to improve the accuracy of meta-object construction, clusters with sufficient differences are generated, and an undirected graph between different clusters is constructed to help achieve accurate layering. And then, with the obtained meta-objects, a complete meta-object representation can be constructed from both the statistical and shape perspectives. Finally, with well-represented meta-objects, popular machine learning techniques can be used for lithology identification tasks. The experimental results show that the feature representation of meta-objects obtained by the proposed ensemble K-means clustering method has a strong reservoir characterization ability. With the well-represented meta-objects, lithology identification accuracy can be significantly improved.