Due to the complexity, risk and specificity of emergencies and the difference and limited knowledge structure of agents, a new alternatives selection method using fusion agents' knowledge level is proposed by introducing large group wisdom into emergency decision-making(EDM). First, the public's knowledge level is determined by preference sequence vectors(PSV) and improved knowledge measurement formulas, and public are grouped by the divergence matrix and controller, and the knowledge level between subgroups is standardized to form a support matrix for the ranking of alternatives based on the public's knowledge level. Second, the experts' knowledge level and attributes are determined by the multi-granularity binary linguistic sets(MG-2LS) and the gray relational analysis(GRA) model. The experts' support matrix for the alternatives ranking is formed by combining the experts' evaluation information obtained by the aggregation operator. Then, the public consensus threshold is introduced to form the consensus matrix which integrates the above-mentioned two agents, and the ranking of schemes is realized. Finally, the validity and rationality of the proposed method are verified by the case of Covid-19 and comparative analysis.