预测导水裂隙带发育高度对煤矿安全开采具有重要意义，在分析类似采矿条件下导水裂隙带发育规律的基础上，选取山东矿区相似地质条件下的36组导水裂隙带发育高度实测数据作为研究对象。以煤层采高、煤层采深、工作面斜长和硬岩岩性比例系数作为导水裂隙带预测模型的主控影响因子，分析其与导水裂隙带发育高度之间的关联性，并基于回归分析和神经网络算法，建立多因素高度关联的导水裂隙带发育高度预测模型。将预测模型的预测值同实测数据和“三下”规范预测值对比分析，结果表明：与导水裂隙带实测值相比，“三下”规范预测值误差绝对值小于5 m的数据占比仅为6 % 和17 %，而回归分析和神经网络2种预测模型的预测值误差绝对值小于5 m的数据占比分别高达83 % 和89 %。2种预测模型的拟合度、稳定性和准确性均优于“三下”规范模型。

Predicting the development height of the hydraulic fracture zone is vital to safe coal mining. This study first analyzed the development patterns of hydraulic fracture zones under similar mining conditions. Taking 36 sets of data measuring the development height of hydraulic fracture zones under similar geological conditions in Shandong mining area as example for analysis, we selected coal seam thickness, mining depth, sloping length of working face, and hard rock lithology ratio coefficient as the main control factors for the prediction model. We analyzed their correlation with the development height of hydraulic fracture zones, and established a multifactorial prediction model with highly-correlated factors via regression analysis and deep learning calculation. Compare and analyze the specification value of the prediction model with the measured data and the "triple down" specification data. Results show that compared with the measured value of hydraulic fracture zones, only 6 % and 17 % of the "triple down" specification data exhibit less than 5m of absolute value of the prediction error, while those in the 2 prediction models via regression analysis and deep learning are 83 % and 89 % respectively. The two prediction models show high curve fit, and their stability and accuracy outperform that of the "triple down" model.