灾害风险判识和预测是控制和治理灾害的前提，工程地质环境是冲击地压等矿井动力灾害最基本的孕灾条件，其沉积成因机制研究对冲击地压风险预测具有重要意义。

以陕西彬长矿区高家堡煤矿为研究对象，分析不同沉积微相下岩体工程地质特性，探究不同沉积环境岩石变形破坏特征及岩体能量释放规律。

结果表明：沉积环境的差异导致岩石岩性和微观结构的差异，岩石加载破坏中能量演化过程大致分为3个阶段，分别为：能量耗散波动阶段、能量耗散平稳阶段、能量耗散阶段。沉积微相的差异导致岩石在能量耗散阶段能量占比变化具有显著差异，河道沉积的细砂岩、中砂岩以及心滩沉积的粗砂岩岩石破坏时能量占比变化较高，在27%以上，而泛滥平原沉积的泥岩岩石破坏时能量占比变化较小，为14%。在此基础上，以沉积微相、煤层厚度、煤层埋藏深度、顶板岩层厚度特征参数、顶板坚硬岩层厚度及其与煤层的间距、岩体质量评估参数、地质构造容量维、侧压系数和弹性能等工程地质环境因素，采用机器学习算法，构建冲击地压风险非线性判识模型。采用BP神经网络、支持向量机(SVM)、决策树(DT)和袋装树(Bagging)等4种机器学习算法进行了对比，各机器学习算法准确率，宏F₁分数以及ROC曲线下面积AUC值均在0.7以上，表明各模型准确度高且稳定性较好。其中，Bagging模型性能最优，表明采用工程地质环境因素能够准确、有效地进行冲击地压风险判识和预测，能够为相似地质条件和开采条件的矿井冲击地压危险性评价提供借鉴，为煤矿防冲卸压设计提供指导和依据。

Disaster risk identification and prediction serve as a prerequisite for disaster control. An engineering geological environment is identified as the fundamental condition inducing dynamic disasters like rock bursts in mines. Exploring the sedimentary genetic mechanisms of engineering geological environments holds great significance for predicting rock burst risks.

With the Gaojiapu Coal Mine in the Binchang mining area, Shaanxi Province, as a case study, this study analyzed the engineering geological characteristics of rock masses under different sedimentary microfacies. Furthermore, it explored the deformation and failure characteristics of rocks in different sedimentary environments, along with the energy release patterns of rock masses.

The results indicate that the differences in sedimentary environments lead to different rock lithologies and microstructures. The energy evolutionary process during rock loading and failure can be roughly divided into three stages: energy dissipation fluctuation, energy dissipation stabilization, and energy dissipation. The differences in sedimentary microfacies result in significantly varying energy proportions of rocks in the energy dissipation stage. Specifically, fine- and medium-grained sandstones deposited in river channels, along with coarse-grained sandstones in mid-channel bars, contribute relatively more energy at rock failure, accounting for more than 24% of the total strain energy. In contrast, mudstones deposited in the flood plains contribute less energy, representing 14%. Accordingly, nonlinear identification models for rock burst risks were constructed using machine learning algorithms, as well as engineering geological environmental factors such as sedimentary facies, the thickness and burial depth of coal seams, the characteristic parameters of rock layer thickness of the roof, the thickness of hard rock layers in the roof and their distance from coal seams, parameters for quality assessment of rock masses, the capacity dimension of geological structures, the coefficient of lateral pressure, and elastic energy. The models built using four machine learning algorithms, namely backpropagation neural network (BPNN), support vector machine (SVM), decision tree (DT), and Bagging, were comparatively analyzed. They yielded accuracy, macro-F1 scores, and area UNDER the receiver operating characteristic (ROC) curve all exceeding 0.7, suggesting their high accuracy and stability. Moreover, the Bagging-based model outperformed the remaining models. The results demonstrate that rock burst risks can be accurately and effectively identified and predicted using engineering geological environmental factors. This study can provide a reference for the risk assessment of rock bursts in coal mines with similar geologic and mining conditions and offer guidance and a basis for the anti-rock burst pressure relief design for coal mines.