矿井全局风流参数实时获取是金属矿井通风系统智能调控的技术关键，目前基于AI算法的风流参数预测方法受数据依赖、计算成本及不同工况适应性的限制，针对该问题，提出了一种适用于金属矿井的全局风流参数智能感知方法。首先，利用风速测量装置实时监测巷道内的平均风速，合理布置监测点，搭建风流参数监测系统，以获取风量、风压等关键通风参数。其次，结合通风系统的实际情况及立体示意图，建立三维仿真系统，并通过现场实测的风流数据对其进行优化，模拟不同通风机运行工况和自然风压状态下的矿井风流流动参数。然后，基于仿真数据，构建了AI算法模型的训练与测试数据集。最后，利用风流参数监测系统采集的风流信息作为 AI 算法模型的输入，实现对矿井全域风流分布的实时感知。利用通风网络解算数据对风流参数智能感知模型进行性能评估，结果表明：① 该模型的决定系数R²为0.998，均方根误差为0.215 9，平均绝对误差为0.085，平均绝对百分比误差为1.89%。② 该模型的预测值与实际观测值之间的吻合度较高，验证了多层感知机(MLP)在风流参数预测方面具备出色性能。③ 该模型在面对不同数据集时能够维持其预测精度，表现出良好的泛化能力。④ 智能通风系统感知数据的平均误差控制在5%以内，井下风流参数感知值与现场实测值基本一致。

Real-time acquisition of global airflow parameters is a key technology for the intelligent control of the ventilation system in metal mines. Currently, AI-based prediction methods for airflow parameters are limited by data dependency, computational costs, and adaptability to different operating conditions. To address this issue, an intelligent perception method for global airflow parameters suitable for metal mines was proposed. First, a wind speed measurement device was used to monitor the average airflow velocity in the roadways in real-time. Monitoring points were strategically arranged, and an airflow parameter monitoring system was established to obtain key ventilation parameters such as air volume and air pressure. Then, based on the actual conditions of the ventilation system and a three-dimensional schematic diagram, a three-dimensional simulation system was developed and optimized using actual measured airflow data. The system simulated the airflow parameters of the mine under different fan operating conditions and natural wind pressure states. Next, based on the simulation data, a training and testing dataset for the AI algorithm model was constructed. Finally, the airflow information collected by the airflow parameter monitoring system was used as input for the AI algorithm model, enabling real-time perception of the global airflow distribution in the mine. Performance evaluation of the intelligent perception model was conducted using ventilation network calculation data. The results showed: ① the model's coefficient of determination (R²) was 0.998, the root mean square error was 0.215 9, the mean absolute error was 0.085, and the mean absolute percentage error was 1.89%. ② The model's predicted values closely aligned with the actual observed values, verifying the excellent performance of the multilayer perceptron (MLP) in airflow parameter prediction. ③ The model maintained its prediction accuracy when faced with different datasets, demonstrating good generalization ability. ④ The average error of the intelligent ventilation system's perception data was controlled within 5%, and the perceived underground airflow parameters were in close agreement with the actual measured values.