Abstract：Aiming at the challenges of data heterogeneity and multi-source in distribution networks，a selfsupervised multi-source measurement data fusion method based on coding-decoding attention mechanism was

proposed. This method automatically captured the correlation between data through self-supervised learning，and

extracted weighted fusion features by encoding and decoding attention mechanisms to enhance the relevance，

integrity and availability of data. This method can adapt to different types of input data，thus ensuring the

realization of high-precision distribution network state estimation in multi-source data scenarios. Experimental

results on a 57-node simulation system show that the proposed method outperforms mainstream algorithms such as GraphMDN，RetNode，AdaAtt and DR-GCN in terms of accuracy，AUC and Macro_F value. Among them，the

accuracy reached 88%，the AUC increased to 76.05%，the Macro_F value reached 93.02%，and the overall

performance was significantly improved. Compared with the optimal comparison algorithm，the average error is

reduced by 47%，and the maximum error is controlled within 0.017. The results verify the effectiveness and

generalization ability of the proposed method in multi-source fusion，power grid data modeling and state estimation.

      Key words：multi-source data fusion；encoding-decoding attention；self-supervised learning；distribution

network state estimation