Considering the low accuracy problem of complex dynamic load identification under the effect of real measurement noise， an L1 norm regularized load identification method based on redundant extended cosine transform dictionary is proposed. Ac? cording to the convolutional relationship between the system response and the external load， the discrete system control equation for load identification is established. According to the main characteristics of the vibration response signal， appropriate discrete co? sine basis functions are selected and extended， and the extended cosine dictionary and the Db10 wavelet dictionary are used to cas? cade a redundant extended dictionary to represent the complex load sparsely. By using the L1 norm regularization method to solve the sparse representation vector under the proposed redundant extended cosine transform dictionary， the optimal regularization pa? rameter is obtained by improved L curve criterion， and the identification of beat load and repetitive impact load at different noise levels is realized. The experimental verification results show that the constructed redundant extended cosine transform dictionary has much better performance in sparse representation of beat load and repetitive impact load， and the load identification method based on the redundant extended cosine transform dictionary has great advantages to obtain accurate inversion results and good ro? bustness.