The vibration signal, frequency components and noise level in multi-sensor measurement of flood discharge structure varies with the measurement position. These signals can be fused to a vibration signal that accurately and comprehensively reflects the vibration characteristics of the overall structure by multi-information fusion technology. As traditional data fusion algorithms fixes convergence factor and requires high similarity for signals, this paper presents the dynamic response data fusion algorithm based on variance dedication rate. Fusion coefficient at a certain time in this algorithm varies with the relative importance of the signal information. The weighting factor is dynamic allocated based on the variance dedication rate. The mathematic signal simulation and prototype measurement of flood discharge structure dynamic response data fusion results showed that dynamic data fusion method proposed in this paper can excavate weak signals hidden in the noise and the fused signal can reflect the vibration characteristics of the overall structure more accurately and comprehensively. The comparison with traditional cross-correlation fusion algorithm shows that this method can effectively highlight the obvious frequency components of the signals, retain the inherent structural frequency components of the original signals and has more applicability.