Abstract: The load identification problem is usually ill-posed, and the structure is often subjected to the combined action of loads at multiple different positions. In order to improve the load positioning accuracy of the structure under multi-source load in the noise environment and improve the peak load identification accuracy of the impact load. In this paper, a multi-source load position identification and reconstruction method based on improved sparse Bayesian is proposed. By introducing a priori of independent accuracy and proposing a parameter update strategy guided by the number of effective parameters \gamma , the position of the load can be automatically identified and the load can be accurately reconstructed. The number of effective parameters can connect the prior distribution and the likelihood function, balance the size of the residual norm and the solution norm, so that the model can adaptively adjust the regularization intensity, thereby achieving sparse Bayesian estimation and accurately estimating the noise level and reconstruction load. In addition, the number of external input loads can be determined by the number of effective parameters, and the identification and time history reconstruction of multi-source load positions are realized. In the cantilever beam test, the average peak error of the proposed method for continuous impact load identification is 0.39 %, which is better than 6.17 % and 5.22 % of the traditional Bayesian method and the fractional order method. Under different noise levels, it still maintains high load identification accuracy and has good noise resistance and robustness. This method is not only superior to the traditional regularization method in the accuracy and stability of the reconstructed load, but also can accurately identify the position of multiple external loads.