为了研究五阳矿采动影响后上覆岩层卸压带高度变化规律，运用BP神经网络预测分析法，以控制“两带”高度的开采方法、工作面长度、工作面推进速度、煤层倾角、覆岩性质、回采高度等致裂因子为输入层，以垮落高度和断裂高度为输出层，借助MATLAB6.x神经网络工具箱，构建了“两带”高度BP神经网络结构，经过BP神经网络模型初始化，BP神经网络模型训练及检验，建立可靠的“两带”高度BP神经网络预测模型。利用BP神经网络预测模型与国外经典预测模型对“两带”高度进行预测对比，得出国外的几种计算方法的误差都比较大，“两带”高度BP神经网络预测模型因其具备非线性函数逼近的特征，计算误差较小；以五阳矿3号煤为研究对象，运用BP神经网络预测模型在多种开采环境下对“两带”高度进行了预测，预测结果与实测结果基本吻合，得出开采方法的改变，影响着“两带”高度变化规律，特定的开采方法条件下“两带”高度与采高呈正相关，增长幅度趋缓；卸采比与采高呈负相关；工作面长与“两带”高度、卸采比均呈正相关，且相关性不受开采方法影响，无论何种开采方法，工作面推进速度与“两带”高度均呈负相关性，且推进速度与“两带”高度线性递减梯度-2.4～4.2 m/(m·d-1)，推进速度增至4 m/d时，“两带”高度几乎不变，表明推进速度是“两带”高度发展的抑制因素。研究结果为五阳矿的瓦斯综合治理提供了重要的技术指导。

In order to study the variation law of the height of the pressure relief zone of the overlying strata after mining in Wuyang Mine, the BP neural network prediction analysis method was used to control the mining method of the “two zones” height. The mining method, working face length, working face advancing speed, coal seam dip angle, overburden property, mining height and other fracturing factors were used as the input layer, and the caving height and fault height were used as the output layer. With the help of MATLAB6.x neural network toolbox, the BP neural network structure of “two zones” height was constructed. After the initialization of BP neural network model, the training and testing of BP neural network model, a reliable BP neural network prediction model of “two iones” height was established. By comparing this model with the classical prediction model abroad,it is concluded that several foreign calculation methods have relatively large errors. The BP neural network prediction model of “two zones” height has the advantages of high prediction accuracy and strong applicability because of its nonlinear function approximation characteristics, which makes up for the defects of other methods. Taking No.3 Coal in Wuyang Coal Mine as the research object, the BP neural network prediction model was used to predict the height of “two zones” under various mining environments. The predicted results are basically consistent with the actual measured results. It is concluded that the change of mining method affects the change law of “two zones” height. Under specific mining method conditions, the height of “two zones” is positively correlated with mining height, and the growth rate tends to slow down; the unloading ratio is negatively correlated with mining height; the length of working face is positively correlated with the height of “two zones” and the unloading ratio, and the correlation is not affected by the mining method. Regardless of the mining method, the advancing speed of the working face is negatively correlated with the height of “two zones”, and the advancing speed and the height of “two zones” decrease linearly with a gradient of -2.4～4.2 m/(m·d-1). When the advancing speed increases to 4 m/d, the height of “two zones” is almost unchanged, indicating that the propulsion speed is the restraining factor of the hight development of “two zones”. The research results provide important technical guidance for comprehensive control of gas in Wuyang Coal Mine.  

0 引言

1 “两带”高度BP神经网络预测模型构建

   1.1 BP神经网络模型结构

   1.2 BP神经网络模型初始化

   1.3 BP神经网络模型训练及检验

2 多模型“两带”高度预测对比研究

   2.1 BP神经网络预测模型适用性分析

   2.2 多模型“两带”高度预测分析

3 五阳煤矿“两带”高度的BP神经网络预测

4 结论