Abstract: Xinglongtai Mesozoic conglomerate oil and gas reservoir is a large integrated tectonic-lithological reservoir discovered in Liaohe Depression in recent years, with low proved reserve and great exploration potential. Oil pool was jointly controlled by lithofacies and tectonicss, the reservoir was highly heterogeneous, and the lithology was complex and diverse, making logging identification difficult and restricting the exploration process in this area. To this end, according to the results of core observation, cuttings logging and thin section identification, the requirements of reservoir classification and evaluation, and the principle of logging identification, the lithology of the study area is re-divided into two categories: clastic rocks and volcanic rocks, including six rock types: granitic conglomerate, mixed conglomerate, sandstone, mudstone, basalt and tuff. A crossplot of lithology was prepared on the basis of analysis of logging response characteristics of different rock types to determine the sensitive logging parameters. On this basis, the K nearest neighbor (KNN) algorithm was improved according to the characteristics of logging data, and the shear proximity machine learning algorithm (MKNN) based on the weighting of logging attributes was proposed and used for lithology prediction. The results showed that compared with the traditional K nearest neighbor (KNN) algorithm, the MKNN algorithm was more efficient, solving the problem that the imbalance of lithology sample types and the overlap of logging parameters, and the accuracy of lithology identification was increased from 82.3% to 88.7%, effectively solving the problem of fine lithology evaluation in old exploration areas. There is much for reference of the study to the improvement of logging evaluation accuracy for complex lithologic reservoirs.

Key words: MKNN algorithm, KNN algorithm, logging response, lithology identification, machine learning, Mesozoic, Liaohe Depression