The multi-agent information fusion(MAIF) system is smainly aimed at information fusion, regulation, communication, and conflict resolution among multiple agents. A multi-agent system conflict data fusion method combining reconstructed basic probability assignment and belief entropy is proposed to address the issue of D-S evidence theory failure under highly conflicting data conditions. This method uses reconstructed basic probability assignment and belief entropy to correct the reliability of evidence, obtaining more reasonable evidence. Then, the evidence is fused using the Dempster combination rule, and the results are obtained with a confidence level of over 90% in both 2 experiments. The experiment demonstrates the effectiveness of this method and improves the accuracy of the MAIF system identification process.