Compared to single atom catalysts (SACs), the introduction of dual atom catalysts (DACs) has a significantly positive effect on improving the efficiency in the electrocatalytic nitrogen reduction reaction (NRR) which provides an environmental alternative to the Haber -Bosch process. However, the research on the mechanism and strategy of designing bimetallic combinations for better performance is still in its early stages. Herein, based on "blocking and rebalance" mechanism, 45 combinations of bimetallic pair doped a -phosphorus carbide (TM A TM B @PC) are investigated as efficient NRR catalysts through density functional theory and machine learning method. After a multi-step screening, the combinations of TiV, TiFe, MnMo, and FeW exhibit highly efficient catalytic performance with significantly lower limiting potentials (-0.17, -0.18, -0.14, and -0.30 V, respectively). Excitingly, the limiting potential for CrMo and CrW combinations is 0 V, which are considered to be extremely suitable for the NRR process. The mechanism of "blocking and rebalance" is revealed by the exploration of charge transfer for phosphorus atoms in electron blocking areas. Moreover, the descriptor u is proposed with machine learning, which provides design strategies and accurate prediction for finding efficient DACs. This work not only offers promising catalysts TM A TM B @PC for NRR process but also provides design strategies by presenting the descriptor u. (c) 2024 Science Press and Dalian Institute of Chemical Physics, Chinese Academy of Sciences. Published by ELSEVIER B.V. and Science Press. All rights are reserved, including those for text and data mining, AI training, and similar technologies.