School of Safety Engineering, North China Institute of Science and Technology
Xi'an Research Institute Co, LTD, China Coal Technology & Engineering Group
Shaanxi Changwu Tingnan Coal Industry Co., LTD.
China Coal Group Shanxi Huayu Energy Co., Ltd.

准确预测导水裂隙带高度是煤炭防治水中的重要问题，也是保障煤矿安全生产的前提。目前，导水裂隙带高度预测模型存在预测准确性差等问题。在广泛收集各地不同煤矿区综采导水裂隙带发育高度实测数据的基础上，总结前人研究成果，选取采厚、工作面斜长、倾角、采深和硬岩比例系数5个因素建立导水裂隙带高度预测模型。同时，采用因素分析法分析了影响因素的权重，确定主控因素，并建立多元线性回归模型、多元非线性回归模型及BP神经网络预测模型，并对预测模型计算导水裂隙带高度进行准确性检验。以西山矿区实测导水裂隙带高度为例，选出最优模型，对最优预测模型BP神经网络模型进行检验分析，并与“三下”规范经验公式、多元线性回归模型和多元非线性回归模型计算结果进行对比。结果表明，“三下”规范经验公式与实际数据存在失真，已不能指导综采条件下导水裂隙带高度预测，而BP神经网络预测模型误差在10%以内，且绝对误差和相对误差较为稳定，同时该模型降低了因素之间的相关性，提高了预测准确性。因此，BP神经网络预测模型在预测导水裂隙带高度方面具有较好的准确性和应用性。研究成果可为指导矿井现场防治水害的工作提供参考建议。

Accurately predicting the height of water-conducting fracture zones is an important issue in coal mine water control and a prerequisite for ensuring mine safety. Currently, there are problems with the accuracy of prediction models for water-conducting fracture zone height. Based on collecting a wide range of measured data on water-conducting fracture zone development height in various coal mining areas and summarizing previous research achievements, a prediction model for water-conducting fracture zone height was established, taking into account five factors: mining thickness, face elongation, dip angle, mining depth, and proportion coefficient of hard rock.Factor analysis was used to analyze the weights of influencing factors and determine the dominant factor. Multiple linear regression model, multiple nonlinear regression model, and backpropagation (BP) neural network prediction model were established. Accuracy tests were conducted on the prediction models to calculate the height of water-conducting fracture zones. Using measured data from the Xishan mining area as an example, the optimal model was selected, and the BP neural network model, as the optimal prediction model, was analyzed and compared with the “Sanxia” empirical formula, multiple linear regression model, and multiple nonlinear regression model in terms of the computed results The results showed that the “San xia” empirical formula had distortions compared to actual data and could no longer guide the prediction of water-conducting fracture zone height under comprehensive mining conditions. On the other hand, the BP neural network prediction model had an error of less than 10%, with stable absolute and relative errors. Additionally, this model reduced the correlation between factors and improved prediction accuracy. Therefore, the BP neural network prediction model demonstrates good accuracy and applicability in predicting water-conducting fracture zone height. The research findings can provide reference suggestions for guiding on-site water hazard prevention and control in mines.

中央高校基本科研业务费资助项目（3142021004，3142020002）；河北省高等学校科学技术基金资助项目（QN2022208）

王　旭，尹尚先，徐　斌，等. 综采工作条件下覆岩导水裂隙带高度预测模型优化[J]. 煤炭科学技术，2023，51（S1）：284−297.

WANG Xu，YIN Shangxian，XU Bin，et al. Study on height optimization prediction model of overburden water-conducting fracture zone under fully mechanized mining[J]. Coal Science and Technology，2023，51（S1）：284−297.