Abstract: As one of the most destructive natural disasters, tornado often cause large casualties and economic losses when it occurs. Due to the unpredictable nature and associated risks of tornado occurrence and development, the current on-site data of tornado are scarce and incomplete. Thusly, this study employed the tornado simulator developed by Wuhan University of Science and Technology to generate the scale tornadoes to investigate the spatial distribution characteristics of three-dimensional wind velocity and pressure drop. Additionally, a fusion model that integrated existing on-site wind-field data and experimental data based on a neural network was proposed to reconstruct the tornado wind field. After validated the model, the tornado-field characteristics under different swirl ratios were predicted. The main findings are as follows: (1) the three-dimensional wind velocity distribution of the simulated tornado vortex in the tornado simulator at Wuhan University of Science and Technology exhibited a remarkable resemblance to that observed in real tornadoes, while the airflow near the core radius demonstrated a spiral state akin to actual tornadoes; (2) the average relative error of tornado wind field and pressure drop predicted by the fusion model was kept below 8%, which had a high precision and could be used for the reconstruction of wind field near the surface and high swirl ratio; (3) the fusion model's prediction results indicated that the maximum tangential wind velocity, vertical wind velocity, and pressure drop value of a tornado were primarily concentrated near the ground, specifically, the maximum tangential wind velocity and vertical wind velocity exhibited concentration near the radius of the vortex core, while the pressure drop value typically reached its peak at the center of the vortex core; (4) the three-dimensional wind velocity and pressure amplitude within the vortex core radius exhibited a significant increase when comparing the reconstructed wind field at a high swirl ratio (S_r=1) to that at a low swirl ratio (S_r=0.17), leading to an expansion of the tornado core area, which encompassed a broader region. This model could completely reconstruct the three-dimensional wind velocity and pressure drop of tornado under different swirl ratios, which provided an important support for the structural wind resistance in the tornado environment.