针对综掘面传统的通风总量控制管理模式不能根据实际需求进行风流调控，造成瓦斯及粉尘聚集和污染隐患等问题，对风流调控下的风速及瓦斯粉尘浓度多源数据融合神经网络预测模型进行了研究。采用欧拉−拉格朗日法建立了风流调控下的瓦斯及粉尘气固耦合模型并进行了测试验证，模拟分析瓦斯和粉尘颗粒在综掘巷道的分布情况，获取大量不同风流调控方案下的风速及瓦斯粉尘浓度样本数据。采用多层感知器神经网络技术建立预测模型结构，选取对瓦斯及粉尘浓度具有较大影响的风流调控等参数作为输入层，根据风速及瓦斯粉尘的隐患位置确定输出层，对样本数据进行预处理，通过引入差分进化算法搜索最佳隐藏层节点数和学习率，利用TensorFlow框架搭建多源数据融合神经网络预测模型。以陕北某矿综掘面为研究对象，对不同风流调控方案进行预测和井下实测验证。结果表明：该模型相对误差最大值为9.7%，具有较高的准确性；选取出风口距端头最短距离5 m和最远距离10 m这2种工况下的最佳调控方案，与调控前相比，风速符合规范要求，端头死角区瓦斯体积分数分别降低34%和35%，回风侧人行处平均粉尘质量浓度分别降低40%和41%，司机处粉尘质量浓度分别降低38%和36%，研究可为风流调控提供参考。

In view of the problem that the traditional ventilation total amount control management mode of the fully mechanized heading face cannot carry out wind flow control according to the actual demand, which causes gas and dust accumulation and pollution hidden danger, the multi-source data fusion neural network prediction model of wind speed, gas and dust concentration under wind flow control is studied. The gas-solid coupling model of gas and dust under wind flow control is established by using the Euler-Lagrange method and tested and verified. The distribution of gas and dust particles in the integrated excavation roadway is simulated and analyzed, and numerous sample data of gas and dust concentration in wind speed under different wind flow control schemes are obtained. Multilayer perceptron neural network technology is used to establish the prediction model structure, and parameters such as wind flow regulation, which have a great impact on gas and dust concentration, are selected as the input layer, and the output layer is determined according to the hidden danger location of gas and dust in wind speed. The sample data is preprocessed, and the differential evolution algorithm is introduced to search the node number and learning rate of the best hidden layer. TensorFlow framework is used to build a multi-source data fusion neural network prediction model. Taking the fully mechanized heading face of a mine in northern Shaanxi as the research object, different wind flow control schemes are predicted and verified by underground measurement. The results show that the maximum relative error of the model is 9.7%, which has high accuracy. The optimal control scheme is selected under the conditions of the shortest distance of 5 m from the outlet and the farthest distance of 10 m from the working face. Compared with before the control, the wind speed meets the standard requirements, the gas concentration in the dead corner of the working face is reduced by 34% and 35%, the average dust concentration at the pedestrian side of the return air side is reduced by 40% and 41%, and the dust concentration at the driver’s side is reduced by 38% and 36%, respectively. The study can provide reference for wind flow control.