The formation of carbonate reservoirs is affected by many factors such as sedimentation, diagenesis, and tectonic evolution. The rock composition and pore structures are complex, which brings challenges to the interpretation of reservoir lithology. Therefore, a novel approach of one-dimensional convolutional neural network architecture (1DCNN) based on the optimization of gradient descent algorithm for lithology identification is proposed. By fully combining logging physical parameters and vertical structure sequence context information, the deeper intrinsic rules between reservoir lithology and wireline logs can be found. With feature extracting of multi-scale deep feature extraction from multivariable wireline logs, the ability of wireline logs to express lithological features is improved, and the accuracy of lithological identification is further improved. In order to illustrate the prediction effect, the carbonate reservoir of the Majiagou Formation in Block 41-33 in Sulige Gas Field is taken as an example. The results reveal that 1DCNN improves the accuracy of carbonate lithology identification by 0.95% to 11.16%, which provides a new idea for the recognition of complex carbonate lithology.