煤巷顶板事故多发生于原生裂隙发育带和岩层破碎带等潜在垮落区，准确、及时掌握顶板岩层地质特征并优化其支护方案及参数是预防顶板事故的重要途径。目前关于巷道顶板岩层地质特征识别方法的相关研究存在识别速度慢、效率低、成本高等问题，难以满足煤矿安全、高效和智能化开采的需求。

提出了一种基于1DCNN(一维卷积神经网络)、BiLSTM(双向长短期记忆网络)和CBAM(卷积块注意力模块)相结合的神经网络模型，建立了随钻振动加速度信号与顶板岩层地质特征的对应关系，通过钻进试验，采集完整岩石、破碎岩石和含裂隙岩石样本的随钻振动加速度信号数据，并将此信号作为训练样本，在此基础上建立了不同岩石样本与其随钻振动信号之间的对应关系，并采用准确率、精确率、召回率和F₁值4项分类指标评估各模型的性能。

研究表明：(1)建立的1DCNN-BiLSTM-CBAM模型实现了端到端的岩层地质特征智能识别，简化识别流程，提高识别效率。(2) SVM(支持向量机)、1DCNN、BiLSTM以及1DCNN-BiLSTM-CBAM模型训练结果表明，1DCNN-BiLSTM-CBAM模型的平均准确率、精确率、召回率和F₁值分别达到99.22%、99.26%、99.21%和99.23%，均优于其他对比模型。(3)实验验证该模型在煤巷顶板岩层地质特征的分类识别中是有效的，且具有良好的分类识别性能，能够满足巷道顶板岩层地质特征实时智能的识别要求。研究为识别煤巷潜在垮落区以及制定顶板支护方案提供了有力的技术支持，提供了一种煤矿巷道顶板随钻探测的高效识别方法，对提升煤矿巷道安全保障技术具有重要的借鉴意义。

Most roof accidents in coal roadways occur in potential caving zones such as primary fissure-bearing zones and rock fracture zones. A significant approach to preventing roof accidents is to understand the geological features of roof strata in an accurate and timely manner and optimize support schemes and parameters of roofs. However, current methods for identifying the geological features of roof strata in roadways suffer from issues such as slow identification speeds, low efficiency, and high costs, thus failing to meet the demand for safe, efficient, and intelligent coal mining.

This study proposed a neural network model based on the 1D convolutional neural network (1DCNN), the bidirectional long short-term memory (BiLSTM), and the convolutional block attention module (CBAM) (also referred to as the 1DCNN-BiLSTM-CBAM model). The correspondence between the acceleration signals of vibrations while drilling and the geological features of roof strata was established. Then, the acceleration signals of vibrations while drilling were acquired from intact, fractured, and fissured rock samples through drilling tests. These signals were used as training samples. Accordingly, the correspondence between different rock samples and the signals of vibrations while drilling was determined. Finally, the performance of various models was assessed using four classification indices: precision, recall, and F₁-score.

The results indicate that the established 1DCNN-BiLSTM-CBAM model allowed for the end-to-end intelligent identification of geological features of rock layers, featuring a simplified identification process and enhanced identification efficiency. The training results of various models revealed that the 1DCNN-BiLSTM-CBAM model exhibited average accuracy, precision, recall, and F₁-score of 99.22%, 99.26%, 99.21%, and 99.23%, respectively, outperforming the support vector machine (SVM), 1DCNN, and BiLSTM models. The experimental verification demonstrates that the 1DCNN-BiLSTM-CBAM model is effective in classifying and identifying the geological features of roof strata in coal roadways, with high classification and identification performance. Therefore, this model can meet the demand for real-time intelligent identification of the geological features of roof strata in roadways. This study presents an efficient identifying-while-drilling method for roadway roofs in coal mines, providing strong technical support for identifying potential caving zones in coal roadways and formulating roof support schemes. This study can serve as a reference for improving the safety guarantee technology for roadways in coal mines.