基于LSTM神经网络的煤矿火灾预测

Coal mine fire prediction based on LSTM neural network

针对传统煤矿火灾预测指标及精度不理想的问题,构建一种具有时序记忆能力的LSTM神经网络模型。利用采空区遗煤氧化过程产生的气体构建模型数据集,通过调整LSTM模型时间步长和迭代次数,分析模型超参数对预测遗煤温度的影响,运用LSTM、GRU和RNN多模型预测分析,验证模型的准确性。结果表明,LSTM神经网络模型比GRU和RNN模型平均绝对误差和平均绝对百分比误差小,分别为3.843、0.029;R^(2)达到0.990,提高了煤矿火灾的预测精度。

This paper features a LSTM neural network model with time series memory skill designed to address the problem that traditional coal mine fire prediction indexes and prediction accuracy are not ideal. The method works by constructing a model data set by using the gas generated in the oxidation process of the goaf leftover coal;analyzing the effect of the model super parameters on the prediction of the leftover coal temperature by adjusting the LSTM model time step and the number of iterations;and verifying the accuracy of the model according to the prediction and analysis of LSTM, GRU, RNN multiple models. The results show that compared with GRU and RNN models, LSTM neural network model has the smallest mean absolute error by 3.843 and mean absolute percentage error by 0.029 respectively;R~2 reaching 0.990 improves the prediction accuracy of coal mine fire. The study provides a reference for the prevention of coal mine fire accidents.