Based on an in-depth study of the strong nonlinear coupling problem in the identification of cable-driven parallel robot systems, this paper proposes an intelligent identification framework that integrates wavelet transforms, Temporal Convolution Network (TCN), and Transformer. This approach first constructs a TCN-Transformer hybrid network architecture to achieve high-precision trajectory prediction and generate virtual data for identification methods to reduce experimental costs. It then separates the high- and low-frequency components of the system’s dynamic characteristics through wavelet multi-scale decomposition. A gray-box identification model is constructed by combining a genetic algorithm with the TCN-Transformer network, enhancing the high-frequency characterization capability while preserving the physical mechanism. Finally, the gray-box identification method identifies the system parameters according to the virtual trajectory generated by the TCN transformer. Experimental verification demonstrates that the proposed method effectively overcomes the shortcomings of traditional identification methods, which suffer from insufficient nonlinear modeling and weak physical interpretability of data-driven methods.