<p>Deep learning has been extensively applied in geophysics, including surface wave exploration. To address issues such as slow convergence, low accuracy, and multiple solutions in surface wave inversion, this study proposes a combined network method (CNN-LSTM) based on convolutional neural networks (CNNs) and long short-term memory (LSTM) to improve the efficiency and accuracy of surface wave inversion. The proposed method first generates fully random velocity structure models and their corresponding dispersion curves to construct a dataset. Then, a combined CNN-LSTM network is constructed to realize nonlinear mapping between dispersion and stratigraphic structure data, thereby enabling direct and rapid inversion of Rayleigh surface waves. To enhance the applicability of the combined CNN-LSTM network, model decomposition techniques are adopted in the absence of prior information, decomposing the original model into a multilayer model to achieve inversion effects for various stratigraphic and velocity structures. Finally, to validate the effectiveness of the proposed method, it is applied to the inversion of surface wave data from the foundation of a comprehensive building of a company in Guilin. The velocity structure indicates that the inversion results are highly consistent with corresponding borehole data, proving the effectiveness of the proposed method and achieving the expected results.</p>