Graph regularized non-negative matrix factorization with optimal basis estimation and its application in recognizing oil reservoir

<p style="text-align:center;"> <span><strong>(Communicated by Changjun Yu)</strong></span> <p style="text-align:justify;"> <span> <span>To solve the issue of low efficiency in interpreting multidimensional unlabeled well logging data, an improved graph regularized non-negative matrix factorization model is proposed. This model takes into account the construction of basis matrix based on low-rank characteristics of the well logging data. Firstly, the rank of basis matrix is estimated by the low-rank matrix recovery model and weighted nuclear norm optimization algorithm. Secondly, the local features of the basis matrix are represented by a spectral clustering on the clean low-rank part. The local features and the optimized low-rank value are integrated to construct the basis matrix, which appropriately reflects the structure that is latent in the data. Finally, the optimized basis matrix is employed and visualized low dimensional non-negative features are extracted for well logging data within the graph regularized non-negative matrix factorization (GNMF) framework. Experimental comparisons on three well logging datasets and their combinations demonstrate the effectiveness of the proposed model, aiding in the efficient and automatic identification of the oil-bearing in reservoirs with unlabeled well logging data.</span><br /> </span>