Porosity is an indispensable key physical parameter in reservoir evaluation, and there exists a complex and potential relationship between well logging curves and porosity. In previous studies, incomplete feature extraction of well logging curves and simple model construction have limited the accuracy of porosity prediction. To improve the prediction accuracy, this study innovatively combines Variational Auto-Encoder (VAE), Bidirectional Gated Recurrent Unit (BiGRU), and Attention mechanism to construct the VAE-BiGRU-Attention model. VAE can effectively learn the latent representation of data, enhancing data representation capability; BiGRU excels at capturing sequential data information, particularly suitable for handling the feature of porosity changing with depth; and the introduction of the Attention mechanism dynamically calculates the attention Attention weights of each time step, allowing the model to more accurately focus on key features and achieve better prediction results. To verify the effectiveness of the model, this paper is compared with Deep Neural Network (DNN), Recurrent Neural Network (RNN), and BiGRU-Attention through comparative experiments. The results show that the VAE-BiGRU-Attention model has a Mean Squared Error (MSE) of 0.995, Mean Absolute Error (MAE) of 0.698, and Root Mean Square Error (RMSE) of 0.998. Compared to other models, it exhibits significant improvement, effectively enhancing the accuracy of porosity prediction and providing a more reliable method for reservoir porosity prediction.