Intelligent matching and lossless migration technology of state grid design achievement documents for project review

Files from different projects often show substantial semantic disparities, requiring an intelligent matching algorithm to enable automatic feature alignment and lossless migration. To overcome the inefficiency and limited reliability of traditional approaches to file matching and migration, this study introduces a file migration model built on multimodal data and a deep convolutional generative adversarial network. The model leverages bidirectional encoding and relational graph convolution to extract textual and image features, applies distance metrics to compute matching scores, and employs hash functions to optimize adversarial training strategies, thereby ensuring lossless feature migration. In experimental evaluations, the proposed model achieved an accuracy of 93.12% in feature matching, an average migration latency of 0.31 ms, and a field fidelity rate of 91.90%. These results indicate that the model delivers more accurate file feature matching and improved migration efficiency, providing algorithmic support for project review outcome file matching.