<p>Lithology identification is a fundamental task in formation evaluation, aiming to recognize and distinguish different lithologies, which is crucial for reservoir exploration and characterization. However, lithology identification faces challenges such as limited well log data samples, imbalanced class distributions, and difficulties in acquiring new samples. Most existing methods attempt to address these issues by generating additional data, but they often fail to fully exploit the relationships between different attributes within the well log data. This paper proposes a lithology identification method based on Generative Adversarial Networks (GANs). The proposed method employs an attribute generator group composed of multiple attribute generators. By cyclically generating sample attributes and using different networks for each attribute, the method avoids the negative impact of sharing network layer parameters when generating different attributes. Additionally, it considers the relationships among various attributes within the well log data. A corresponding pseudo-labeling mechanism is designed to process the generated data, ensuring that the classifier is trained automatically with data that has high positive utility. Compared to traditional deep learning methods, this approach achieves an average accuracy improvement of 8.23 percentage points on a U.S. dataset.</p>