Precise simulation of weak structural surfaces in geomechanical model test is critical to their success. This study addresses a key challenge in geotechnical testing: the difficulty of selecting and configuring similar materials that accurately replicate shear strength parameters. An indirect inversion method and process is introduced for determining similar material proportions based on improved adaptive genetic algorithm–backpropagation (IAGA-BP) neural network, which significantly enhances inversion accuracy and concurrently reduces the number of experimental samples required. Talcum powder, sand, Vaseline, and oil are used to configure similar materials, with a focus on simulating the shear strength parameters f and c for weak structural surfaces in the model test. A total of 168 direct shear tests were conducted across various material proportions and normal stresses, producing 168 shear strength datasets and 42 friction coefficient and cohesion datasets. The selected material and configuration method allow for a wide range of shear strength parameters, with a friction coefficient of 0.264–0.687 and cohesion of 0.03–20.66 kPa. In the proposed indirect inversion method, the BP neural network is employed to indirectly predict the shear strength instead of friction coefficient and cohesion. Subsequently, the f and c values are fitted by Mohr-Coulomb criterion based on the predicted shear strength data. Finally, the IAGA algorithm is used to search for the optimal mix ratio close to the target value. To determine the applicability and robustness of the proposed mix ratio determination method, a systematic study is conducted to evaluate the performance of the indirect and traditional inversion methods, investigate the impact of the sample size on prediction accuracy, and analyze the influence of various factors on f and c. The test results demonstrate that, even with a limited number of experiments, the indirect inversion method achieves higher accuracy than the traditional inversion method in mix ratio determination.