The application of dictionary learning for moving force identification (MFI) is utilized to analyze complex and unknown moving vehicle loads. However, existing dictionary learning technologies have limited capacity in learning non-sparse forces. To address this issue, a novel multi-domain dictionary learning model integrating time and frequency domain representations is proposed to enhance the applicability and accuracy of MFI. Firstly, the moving force is decomposed into narrowband and broadband force components in the frequency domain, where the former can be sparsely represented in the dictionary, while the latter does not exhibit sparse characteristics. Subsequently, the dictionaries are constructed in both frequency and time domains to represent the narrowband and broadband force components, respectively. The double sparse coding technique is employed to update the frequency domain dictionary and its atomic coefficients, and the elastic network regression technique is introduced to update the time domain dictionary, which can be more flexible to accommodate different types of broadband force components. A series of numerical simulations are carried out to assess the effectiveness of the proposed method. The results show that the proposed multi-domain model significantly outperforms existing single-domain method, especially in the identification of moving random forces with higher accuracy and noise robustness. The integration of time-frequency representations within a multi-domain dictionary learning framework proves to be a powerful and effective solution to fix the MFI problem in practice.