The frequent occurrence of broadband oscillation issues in "dual-high" power system poses a significant threat to its secure and stable operation. Extracting broadband oscillation modal information through identification techniques provides critical characteristic parameters for suppression measures. To address this issue, this paper proposes a broadband oscillation modal parameter identification method based on improved variational mode decomposition-exponentially damped sinusoids neural network (IVMD-EDSNN). To address the challenge that existing signal processing methods struggle to accurately separate nonlinear broadband oscillation signals, a windowed interpolated fast Fourier transform and dispersion entropy are introduced to improve variational mode decomposition (VMD), which lacks adaptive parameter selection. The original broadband oscillation signal is decomposed into multiple single-frequency modal components using IVMD. Subsequently, an EDSNN model is constructed with a topological structure that aligns with the mathematical model of broadband oscillation signals. By integrating the IVMD decomposition results, the method enables precise identification of the modal parameters of broadband oscillation signals. The effectiveness of the proposed approach is validated through numerical broadband oscillation signal analysis and MATLAB/Simulink simulation tests. Comparative experiments with the Prony algorithm and the total least squares-estimation of signal parameters via rotational invariance techniques (TLS-ESPRIT) algorithm demonstrate that the proposed method achieves higher accuracy in identifying the modal parameters of broadband oscillation signals.