Abstract: Firstly, a variable scale integral function was formed based on the idea of the integral moving average, and a weighted integral moving average optimal approximate response function model was proposed under least squares conditions. Secondly, a shape function method was used to construct optimal approximate load model. The real Duhamel integral equation can be obtained with the combination of approximate load and response. The discretization of Duhamel integral results in a systematically discretized equation, which can be used for load identification. Thirdly, the regularization method was used for load identification, by using the least squares solution of the regularization method to construct a function of the regularization parameter. A new method was proposed for choosing the optimal regularization parameter. Finally, numerical examples and experimental verification illustrated that the new method can obtain a stable solution with better accuracy and noise resistance.