A method for identifying diagenetic facies in tight sandstone reservoirs has been proposed, addressing the challenges of low porosity, low permeability, and strong heterogeneity that lead to low precision in diagenetic facies logging identification, difficulty in identifying thin layers, and the reliance on manual processes in traditional methods. The proposed method utilizes a fusion of multi-scale convolutional gated recurrent units, based on gated recurrent units, combined with depthwise separable convolutions to integrate features and historical states. A multi-scale fusion mechanism is introduced to enhance the extraction of diagenetic facies features at different thicknesses. Channel attention and spatial attention are employed to boost the response to key features.The study takes the Fuyu oil layer in the Sanzhao depression of the Songliao Basin as the research object, constructing a six-class diagenetic facies dataset and conducting validation experiments. The results demonstrate that the proposed method achieves an F1 score and balanced accuracy of 0.92, maintaining high recognition accuracy under noise interference. The method exhibits optimal runtime performance under different data volumes, with an individual well recognition accuracy of 86.3%. Its overall performance surpasses that of models such as random forests, convolutional neural networks, and gated recurrent units. This method realizes the collaborative modeling of spatial and temporal features of logging data, effectively improving the accuracy, robustness, and efficiency of diagenetic facies recognition, providing a new technical pathway for intelligent recognition of diagenetic facies in tight sandstone reservoirs.