基于遗传算法-深度神经网络的分布式光纤监测工作面矿压预测

煤层开采过程中的工作面矿压分析与预测，对煤矿顶板管理与安全生产具有重要意义。然而，工作面开采引起的围岩移动和变形影响着矿压预测的准确度。为了提高工作面来压位置预测的精度，以分布式光纤监测采动覆岩变形的频移数据为基础，引入门控循环神经网络(gated recurrent neural networks, GRU),建立了遗传算法(genetic algorithm, GA)-GRU-反向传播(back propagation, BP)的工作面来压位置预测模型。将光纤频移值的统计特征融合工作面推进距离等因素作为特征向量，并采用GA对GRU及BP网络的超参数寻优。实验结果表明：预测模型的决定系数为98.7%,平均绝对误差为1.224 cm,均方根误差为1.769 cm,预测的准确性高，为工作面矿压预测提供了新的方法。

Mine Pressure Prediction of Distributed Optical Fiber Monitoring Based on GA-Deep Neural Network on Working Face

In order to obtain the regulation s of mining pressure and predict the location of the pressure on the working face accurately, a physical simulation test of similar materials was set up, and distributed optical fibers were used to monitor the deformation of the overlying rock during coal mining. Using gated recurrent neural networks, Combining the statistical characteristics of the frequency shift value of the optical fiber sensor and other factors affecting the rock pressure, such as the location of working face, the GA-GRU-BP position prediction model is established on working face. Using Genetic Algorithm (GA) to optimize the hyperparameters during the GRU network training. Correlation coefficient (R2), mean absolute error (MAE) and root mean square error (RMSE) are used to evaluate algorithm performance. GA-GRU-BP is compared with random forest (RF), GA-support vector machine(SVM), and GA-GRU. The experimental results show that the R2, MAE, and RMSE of the GA-GRU-BP prediction model are98.7%, 1.224cm, and 1.769cm, respectively, which are all lower than the evaluation indicators of the corresponding comparison method. The GA-GRU-BP position prediction model has a higher accuracy and robustness. A new method for predicting the position of the mining pressure is provided on working face.