我国华北型煤田煤炭资源开采深度及开采强度逐渐增加，进而导致覆岩移动破坏及裂隙演化等热点问题，引起极大关注。

收集华北型煤田导水裂隙带高度实测数据共117组，其中不同覆岩类型(坚硬、中硬和软弱)数据分别为17、42和58组。在此基础上探究不同覆岩类型控制下不同采高、不同采深以及不同工作面斜长对导水裂隙带发育高度的影响。根据华北型煤田含煤地层的沉积特点，将其划分为北带、中带和南带3个区域，分别采用卷积神经网络、贝叶斯公式及《“三下”规范》3种方法，分区域开展北带、中带和南带地区的导水裂隙带高度精细化研究。

结果表明：随着采高、采深或工作面斜长等因素的改变，不同覆岩类型条件下的导水裂隙带高度分布出现明显差异，从坚硬、中硬到软弱覆岩，裂采比依次降低，其中坚硬覆岩的裂采比是中硬覆岩的1.59倍，是软弱覆岩的1.77倍；中硬覆岩的裂采比是软弱覆岩的1.11倍。预测结果显示，3个地区卷积神经网络和贝叶斯公式的RMSE(E_RMS)值分别为6.62和21.84、2.20和8.09、2.60和6.12，明显小于《“三下”规范》中的经验公式的RMSE(E_RMS)值(45.91、13.40和21.99)，说明卷积神经网络和贝叶斯公式的预测结果均优于经验公式，其中卷积神经网络预测结果更为贴近实测结果，具有良好的适用性，可为华北型煤田不同覆岩类型下的导水裂隙带高度预测提供依据。

The gradual increase in the exploitation depth and intensity of coal resources in the North China-type coalfields has caused problems such as overburden movement and damage, as well as fracture evolution. These hot topics have attracted considerable attention.

A total of 117 sets of measured data on the heights of water-conducting fracture zones in the North China-type coalfields were collected, consisting of 17 sets from hard overburden, 42 sets from moderately hard overburden, and 58 sets from soft overburden. Using these data, this study explored the effects of varying mining heights, mining depths, and lengths of the mining face along its dip direction on the heights of water-conducting fracture zones under varying overburden types. Given the sedimentary characteristics of their coal-bearing strata, the North China-type coalfields were divided into three regions: the northern, middle, and southern belts. This study thoroughly investigated the heights of water-conducting fracture zones in the three regions using the convolutional neural network, the Bayes' formula, and empirical formulas for coal mining under buildings, water bodies, and railways.

The results indicate that as the mining height, mining depth, or the length of the mining face along its dip direction varied, the height distributions of water-conducting fracture zones differed significantly under different overburden types. From hard, to medium-hard, and then to soft overburden types, the ratio of the height of water-conducting fracture zones to the mining height (also referred to as the fracture-to-mining ratio) decreased sequentially. Specifically, the fracture-to-mining ratio of hard overburden was 1.59 times that of moderately hard overburden and 1.77 times that of soft overburden, while the fracture-to-mining ratio of moderately hard overburden was 1.11 times that of soft overburden. The prediction results indicate that the prediction results of the northern, middle, and southern belts calculated using the convolutional neural network yielded root mean square errors (RMSEs) of 6.62, 2.20, and 2.60, respectively, while those calculated using Bayes’ formula yielded RMSEs of 21.84, 8.09, and 6.12, respectively. These values were much less than those derived using the empirical formulas for coal mining under buildings, water bodies, and railways (45.91, 13.40, and 21.99, respectively). This suggests that the convolutional neural network and Bayes’ formula outperform the empirical formulas. Notably, the prediction results obtained using the convolutional neural network are closer to the measured results, suggesting the high suitability of the convolutional neural network. This study can provide a basis for predicting the heights of water-conducting fracture zones under different overburden types in the North China-type coalfields.