China Coal Research Institute
CCRI Tong'an (Beijing) Intelligent Control Technology Co., Ltd.
Beijing Engineering and Research Center of Mine Safe
State Key Laboratory of Intelligent Coal Mining and Stratum Control

针对综掘工作面降尘措施难以合理利用的问题，提出了一种基于本征正交分解（POD）与机器学习的综掘工作面流场快速预测算法。利用计算流体力学（CFD）技术对多种工况下的综掘工作面风流场和粉尘浓度场进行模拟，获得高维度流场数据；利用POD对高维度流场数据进行降维，提取能够反映流场主要特性的核心模态，得到流场工况的基函数模态与模态系数；通过机器学习方法预测不同工况下占总能量达到90%以上的模态系数，从而对未知工况的模态系数进行预测，利用预测的模态系数与基函数模态进行重构，得到未知工况的风流场数据或粉尘浓度场。研究结果表明：综掘工作面数值模拟模型的相对误差在3%以内，能够准确反映实际的风流和粉尘分布状况；风流场选择前5阶模态，粉尘浓度场选择前7阶模态，即可兼顾POD重构精度与效率；支持向量机（SVM）模型对模态系数的预测能力优于随机森林及神经网络模型，针对60种不同工况，基于POD和SVM预测的风流速度、粉尘浓度与数值模拟结果之间的相对误差分别为0.36 m/s，86.24 mg/m3，风流场及粉尘浓度场平均预测耗时为73 s，实现了对矿井综掘工作面风流场和粉尘浓度场的高精度快速预测。

To effectively utilize dust suppression measures in fully mechanized excavation faces, this study proposed a rapid prediction algorithm for the flow field based on proper orthogonal decomposition (POD) and machine learning. First, computational fluid dynamics (CFD) technology was used to simulate the air flow field and dust concentration field under various conditions, generating high-dimensional flow field data. Then, the POD method was applied to reduce the dimensionality of this data, extracting core modes that captured the main characteristics of the flow field and producing basis function modes and mode coefficients. Machine learning techniques were subsequently used to predict the mode coefficients that accounted for over 90% of the total energy under different conditions, enabling predictions of mode coefficients for unknown conditions. Finally, by reconstructing the flow or dust concentration field data using the predicted mode coefficients and basis function modes, rapid and accurate predictions for the flow field in excavation faces were achieved. The results showed that the numerical simulation model for the excavation face had a relative error within 3%, accurately reflecting the actual air flow and dust distribution. Selecting the first five modes for the flow field and the first seven modes for the dust concentration field balanced the accuracy and efficiency of POD reconstruction. The support vector machine (SVM) model outperformed the Random Forest and Neural Network models in predicting mode coefficients. For 60 different conditions, the relative errors between the POD and SVM-predicted flow velocity and dust concentration, and the CFD results, were 0.36 m/s and 86.24 mg/m³, respectively. The average prediction time for the flow and dust concentration fields was 73 seconds, achieving high-precision, rapid predictions for airflow and dust concentration in mine excavation faces.

国家重点研发计划项目（2023YFC3009001-8）；天地科技科技创新专项重点项目(2024-TD-ZD011-01)；煤科院技术创新基金项目（2021CX-II-15）。

金兵，张浪，李伟，等. 基于POD与机器学习的综掘工作面流场快速预测算法[J]. 工矿自动化，2024，50（10）：97-104, 119.

JIN Bing, ZHANG Lang, LI Wei, et al. Rapid prediction algorithm for flow field in fully mechanized excavation face based on POD and machine learning[J]. Journal of Mine Automation，2024，50（10）：97-104, 119.