The traditional model updating technique uses intelligent algorithms to find a set of global optimal solutions，which makes the finite element model closer to the actual structural response. Actually，since the number of objective functions is often smaller than the number of parameters to be optimized，there may be more than one global optimal solution. Because the number of sensors is limited，the measurement data is insufficient，and the instrument identification has errors，the local optimal solution may represent the actual structural parameters more than the global optimal solution. Based on that，the paper proposes a niche fish swarm algorithm，which introduces niche technology and reverse learning strategy，and uses simulated annealing algorithm to perform detailed optimization in the later stage of the algorithm，It overcomes the drawbacks of the artificial fish swarm algorithm that it is difficult to find multiple function peaks，sensitive to initial values，slow convergence and low precision，and it is verified by numerical simulation to apply it to a scaled cablestayed bridge model. In order to avoid a large number of calls to the finite element model，this paper uses the response surface as a proxy model，generates experimental data through the Latin hypercube，performs a significant test on each factor，and eliminates the unrelated factors to test the fitting accuracy. The model updating results in a set of global optimal solutions and 7 sets of local optimal solutions that reduce the error between the calculated and measured values of the scaled cable-stayed bridge，Thus

decision makers can choose from multiple sets of solutions based on engineering experience

reducing the possibility of misjudgement of actual structural parameters.