<p>Logging data is a critical resource in geological exploration. However, during actual logging operations, data loss, anomalies, or distortions often occur due to instrumentation issues or geological conditions, posing challenges for subsequent geological interpretation and resource evaluation. This paper proposes a novel method for reconstructing logging curves based on Characterisation-enhanced Variational Autoencoder (CEVAE). This method extends the traditional variational autoencoder (VAE) by integrating logging geological prior knowledge, proposing an attribute feature cross strategy, and introducing the Efficient Channel Attention mechanism (ECA) along with the PlanarFlow standardised flow to strengthen the model’s capacity to represent latent variables in logging data. Additionally, we design an adaptive encoder and decoder architecture to further improve representation learning. Our proposed model performs well on real uranium mine logging datasets, demonstrating the effectiveness and superiority of the CEVAE model in addressing missing value issues from a multi-attribute perspective, and significantly improving reconstruction accuracy.</p>