Shaanxi Binchang Hujiahe Mining Co., Ltd.
Shaanxi Coal and Chemical Industry Technology Research Institute Co., Ltd.
College of Energy Engineering, Xi'an University of Science and Technology

现有工作面矿压显现规律预测方法中，基于数值模拟与统计回归的方法无法实现对工作面矿压显现规律的实时精准预测，深度学习方法存在超参数较多且难以设置、模型训练速度慢等问题。针对上述问题，以胡家河煤矿402102回采工作面采动过程中监测到的煤体内部应力变化时序数据为基础，将基于粒子群优化的门控循环单元（PSO−GRU）应用到回采工作面矿压显现规律预测中。采用PSO算法对GRU进行优化，构建PSO−GRU模型，实现对超参数的自动寻优，从而提高GRU的训练速度和预测精度。以预测结果为依据，采用层次分析法建立402102回采工作面矿压主控因素评价指标体系，将顶板条件、回采工艺、煤层赋存、地质构造确定为影响工作面矿压的一级指标，进一步细分出具有代表性的14个二级指标。测试结果表明：① 与未经优化的GRU模型相比，PSO−GRU模型的均方误差（MSE）降低了83.9%，均方根误差（RMSE）降低了59.8%，平均绝对误差（MAE）降低了59.0%，决定系数R2提升了28.9%。② PSO−GRU模型对矿压数据预测的拟合度达0.980以上，具有良好的非线性拟合能力和泛化能力。③ 地质条件中的煤层赋存因素对回采工作面矿压的影响最大，权重为0.47；可人为干预的影响因素中工作面推进速度对矿压的影响最大，权重为0.13。

Among the existing prediction methods for strata behaviors law in working faces, the methods based on numerical simulation and statistical regression cannot achieve real-time and precise prediction of strata behaviors law in working faces. Deep learning methods have problems such as a large number of hyperparameters that are difficult to set and slow model training speed. In order to solve the above problems, based on the time-series data of internal stress changes in the coal body monitored during the mining process of the 402102 working face in Hujiahe Coal Mine, the particle swarm optimization based gate recurrent unit (PSO-GRU) is applied to predict the strata behaviors law in the working face. The PSO algorithm is used to optimize GRU. The PSO-GRU model is constructed to achieve automatic optimization of hyperparameters, thereby improving the training speed and prediction precision of GRU. Based on the prediction results, the analytic hierarchy process (AHP) is used to establish the evaluation index system of the main control factors of the strata behaviors of the 402102 mining face. The roof conditions, mining technology, coal seam occurrence, and geological structure are identified as the first level indicators affecting the strata behaviors of the working face. The 14 representative second level indicators are further subdivided. The test results show the following points. ① Compared with the unoptimized GRU model, the mean square error (MSE) of the PSO-GRU model is reduced by 83.9%, the root mean square error (RMSE) is reduced by 59.8%, the mean absolute error (MAE) is reduced by 59.0%, and the coefficient of determination R2 is increased by 28.9%. ② The PSO-GRU model has a fitting degree of over 0.980 for predicting strata behaviors data, demonstrating good nonlinear fitting and generalization capabilities. ③ The occurrence factors of coal seams in geological conditions have the greatest impact on the strata behaviors of the mining face, with a weight of 0.47. Among the factors that can be intervened by humans, the impact of advancing speed on the strata behaviors of the working face is the greatest, with a weight of 0.13.

国家自然科学基金面上项目（52074209）；陕西省自然科学基础研究计划联合基金项目（2021JLM-06）。

席国军，余智秘，李亮，等. 胡家河煤矿综放工作面矿压显现规律预测及主控因素研究[J]. 工矿自动化，2024，50（1）：138-146.

XI Guojun, YU Zhimi, LI Liang, et al. Prediction of mine strata behaviors law and main control factors in the fully mechanized caving face of Hujiahe Coal Mine[J]. Journal of Mine Automation，2024，50（1）：138-146.