With the rapid development of smart grids，traditional methods of power data transmission and data access sharing are increasingly inadequate in ensuring the privacy，security，and efficiency of power data.Therefore，this study proposes a cross-domain model for power big data based on secure federated learning.Initially，the study analyzes the specific architecture and existing challenges of the cross-domain joint model for power big data.To achieve secure aggregation of cross-domain data in a distributed environment，an integrated federated learning framework is employed to construct a cross-domain model with privacy protection.The results demonstrate that the leakage rate of shared data in the proposed model is less than 1%，and it takes approximately 2 seconds to transmit 100 data copies.The computational cost is significantly lower than that of traditional models.In practical applications，the relative error between the cross-domain aggregated data for model training and the standard data is less than 0.5%.The experimental data indicate that the model successfully balances privacy security，recommendation accuracy，and data authenticity，enabling secure and accurate data sharing.This advancement is expected to promote the development and application of cross-domain information sharing technology.