致灾机理、危险性评价、灾变预测共同构成矿井水防治理论体系基本内容，其在过去20多年里快速发展，目标是理解矿井水行为特征，预测演化趋势，服务矿区水害防治工作。机器学习是大数据时代进行数据分析和挖掘的有力工具。将机器学习应用于矿井水防治理论体系研究，已得到相对广泛的关注。针对理论体系的3项基本内容，重点讨论了机器学习在各内容建设中的具体应用，主要包括：根据不同水害类型分类简述致灾机理研究现状，指出机器学习应用暂为空白的原因为其不具备做出假设的能力。认为未来致灾机理研究方法依然以传统方法(理论分析、数值模拟、相似模拟等)为主，机器学习促进地质数据获取与处理，对机理研究作出贡献；分析方法优势，指出机器学习作用于危险性评价的主要方式为非结构化数据的处理及丰富评价方法；分析基于物理和基于数据的单一预测模式弊端，论述物理模型与数据驱动相结合的必要性，相应给出“模型−数据”双驱动预测模式的3种实现形式，并讨论了基于图像的灾变预测方法可行性。随着生产数据及地质数据的丰富，机器学习方法可推动理论体系研究快速发展，并为矿井水防治学科系统方法论研究作出贡献。

The theoretical system of mine water prevention and control encompasses three fundamental aspects: disaster-causing mechanisms, risk evaluation, and disaster prediction. This theoretical system, having undergone rapid development over the past 20 years, aims to gain insights into the behavior characteristics of mine water and predict its evolutionary trend, thus serving the prevention and control of water disasters in mining areas. Applying machine learning, a powerful tool for data analysis and mining in the era of big data, to research into the theoretical system has garnered considerable attention. This study focuses on the specific applications of machine learning to the three fundamental aspects of the theoretical system. Specifically, this study offered a brief introduction to the current status of research on disaster-causing mechanisms based on the classification of varying water disasters, proposing that the application gap of machine learning to the mechanism research is due to its incapacity to make assumptions. This study posited that future research on disaster-causing mechanisms will still primarily rely on conventional methods like theoretical analysis, numerical simulation, and similarity simulation, with machine learning facilitating the acquisition and processing of geologic data. The analysis of method advantages reveals that the application of machine learning to the risk evaluation primarily via processing unstructured data and enriching evaluation methods. For disaster prediction, this study analyzed the drawbacks of prediction modes based merely on physics or data and expounded on the necessity of combining physical models with data-driven approaches. Accordingly, this study presented three methods for achieving the model-data dual-driven prediction mode. Additionally, this study explored the feasibility of image-based disaster prediction methods. With the increasing abundance of production and geologic data, machine learning will accelerate the development of the theoretical system, contributing to research on the systematic methodology for mine water prevention and control.