In order to realize the adaptive decomposition and reconstruction for the response signals of bridge structure as well as the intelligent identification of mode parameters, a novel corresponding improved method is proposed by aiming at the existing ensemble empirical mode decomposition(EEMD)and intrinsic mode function for the problems of existing end effects and selecting effective modal functions difficultly. Based on the introducing the mirror extension method and the support vector regression algorithm, the endpoint effects is suppressed. And at the same time, in accordance with the cross correlation coefficient and energy coefficient, a new index of effective intrinsic mode function is established. Moreover, the algorithm for intelligent identification of the real modal state is proposed based on the general rule of the true mode of bridge structure. The effectiveness of the proposed method is verified by a shaking table test for a model of cable-stayed bridge. The verified results showes that the proposed method can not only realize adaptive decomposition and reconstruction of response signals for bridge structural, but also can implement intelligent screening of the real modal in the stable graph, that is realize the intelligent identification of modal parameter for the bridge structural, and its recognition results are reliable.