Abstract: The coal seam in Yushenfu mining area is shallow, the overlying bedrock is thin, and the loose layer is thick. Most mines involve repeated mining of multiple coal seams. Affected by multiple factors such as coal seam mining thickness and spacing, the spatial interaction of surrounding rock in the upper and lower stopes makes it difficult to accurately predict fracture zone height. In this paper, the height of multi-coal seam repeated mining fracture zone in typical coal mines in Yushenfu mining area is taken as the research object. The research methods of physical similarity simulation, theoretical analysis and deep learning are used to analyze the development law of multi-coal seam repeated mining fracture. The multi-factor coupling nonlinear regression model of coal seam mining thickness, spacing, burial depth, dip angle, working face length and interval rock strength and fracture zone height is constructed. The prediction method of multi-coal seam repeated mining fracture zone height based on SSA-BP neural network is established, and its accuracy is verified. The results indicated that the fracture development under repeated mining exhibited a three-stage characteristic: "localized slow growth - nonlinear rapid increase through interconnection - dynamic stabilization." The ultimate height of the fractured zone reached 139.0 m. A nonlinear regression model incorporating coal seam mining height, interlayer spacing, strength of intervening rock strata, and working face length achieved an R² value of 0.880, confirming these parameters as key influencing factors for the development height of the fractured zone. Compared to predictions from traditional empirical formulas and the BP model, the SSA-BP model demonstrated reductions in MAPE values by 22.96% and 6.70%, respectively, with an RMSE as low as 1.79, indicating superior stability. Validation at the 14205 working face of the Zhonghui Funeng Coal Mine in the Yushenfu Mining Area showed a relative error between the predicted and measured heights of 1.3%, which was less than 5%. The study demonstrated strong generalizability for predicting the height of water-conducting fracture zones in multi-coal seam mining in the Yushenfu Mining Area, providing valuable insights for water hazard prevention and control in such mining conditions.