榆神矿区中厚煤层赋存量大、地质条件相对简单，布置超长工作面是提高工作面产效、减少煤炭损失的有效途径之一。工作面倾向长度增加使得综采工作面矿压显现程度和空间分布特征异于一般工作面，给超长工作面顶板管理与智能控制带来较大难题。因此，分析超长工作面支架工作阻力及来压特征并借助深度学习算法实现工作面来压精准预测，对于保障中厚煤层超长工作面安全智能高效开采具有重要意义。

本文以榆神矿区中厚煤层超长工作面开采过程为研究对象，采用理论分析、现场实测、深度学习等综合研究方法，分析了超长工作面支架工作阻力区域分布特征，提出了一种基于自适应图卷积循环网络的超长工作面来压预测模型，开发了超长工作面来压实时监测及预警系统。主要研究成果如下：

根据榆神矿区典型超长综采工作面地质与开采条件，通过理论分析和现场实测，分析了支架工作阻力分布特征和来压规律。通过关键层判定、砌体梁理论模型计算及支架工作阻力分布特征实测，确定了超长工作面来压步距；超长工作面支架工作阻力沿倾向整体具有“低-高-中-高-低”分布特征，工作面上、下端部区域支架工作阻力普遍较小，中部支架工作阻力较大。

（2）结合图理论，运用自适应图卷积循环网络（AGCRN）模型提取工作面支架工作阻力数据之间的时间及空间关联关系，建立了基于AGCRN算法的超长工作面来压预测模型。对比BP模型、GRU模型及DCRNN三个基准模型，AGCRN预测模型在测试集上的MAE和MAPE值最小，说明AGCRN模型在工作面来压预测时具有良好的性能。

（3）基于建立的AGCRN来压预测模型，利用Python开发语言和MySQL数据库，结合Django框架和Vue框架开发了中厚煤层超长工作面来压预警系统，该系统主要包括了系统管理、支架工作阻力实时监测与反馈和来压预测等模块，为中厚煤层超长工作面支架状态监测、来压识别与智能控制提供了基础依据。

The medium-thickness coal seam in Yushen mining area has a large amount of occurrence and relatively simple geological conditions. The arrangement of ultra-long working face is one of the effective ways to improve the production efficiency of working face and reduce coal loss. The increase of length of working face makes the degree and spatial distribution characteristics of mine pressure in fully mechanized working face different from those in general working face, which brings great difficulties to roof management and intelligent control of ultra-long working face. Therefore, it is of great significance to study the pressure law of ultra-long working face and accurately predict the mine pressure behavior of working face with the help of deep learning algorithm for realizing safe, intelligent and efficient mining of ultra-long working face in medium-thickness coal seam.

In this paper, the mining process of ultra-long working face in medium-thickness coal seam of Yushen mining area is taken as the research object. The distribution characteristics of support working resistance area of ultra-long working face are analyzed by theoretical analysis, field measurement, deep learning and other comprehensive research methods. A pressure prediction model of ultra-long working face based on adaptive graph convolution cycle network is proposed, and a mine pressure monitoring and early warning system for ultra-long working face is developed. The main research results are as follows:

(1) According to the engineering geological and mining conditions of typical ultra-long fully mechanized mining face in Yushen mining area, through theoretical analysis and field measurement, the distribution characteristics of working resistance and pressure appearance law of support are analyzed. Based on the distribution characteristics of the working resistance of the support in the ultra-long working face along the strike direction, the theoretical calculation value of the pressure step distance in the ultra-long working face is basically consistent with the measured value. The working resistance of the support has zoning characteristics along the inclined direction. The working resistance of the support is small in the upper and lower areas of the working surface, and the middle support is larger. There is a ' bimodal ' distribution along the inclined direction of working face. The working resistance of the support has the distribution characteristics of ' low-high-medium-high-low '.

(2) Combined with the graph theory, the adaptive graph convolution recurrent network ( AGCRN ) model is used to extract the temporal and spatial correlation between the working resistance data of the working face support, a pressure prediction model of ultra-long working face based on AGCRN algorithm is established. The mean absolute error (MAE) and mean absolute percentage error (MAPE) are selected as indicators. Compared with the three benchmark models of BP model, GRU model and DCRNN, the MAE and MAPE values of the AGCRN prediction model on the test set are the smallest, indicating that the AGCRN model has good performance in the pressure prediction of ultra-long working face.

(3) Based on the established AGCRN pressure prediction model, using Python development language and MySQL database, combined with Django framework and Vue framework, a pressure early warning system for ultra-long working face in medium-thick coal seam is developed. The system mainly includes system management, real-time monitoring of working resistance of working face support and feedback, pressure prediction modules, which provides a basis for support condition monitoring, pressure identification and intelligent control of ultra-long working face in medium-thick coal seam.