基于深度神经网络的煤矿瓦斯浓度序列预测算法

Coal mine gas concentration sequence prediction algorithm based on deep neural network

针对统计学习和机器学习方法难以对瓦斯浓度序列数据准确预测的问题,提出一种基于长短期记忆(LSTM)-门控循环单元(GRU)神经网络的瓦斯浓度序列预测算法。首先对数据进行划分和归一化;接着引入LSTM神经网络细胞和GRU神经网络细胞处理具有时序性的历史瓦斯浓度序列数据,设计网络结构学习瓦斯浓度序列内部动态变化规律,以误差损失最小化为目标,得到预测方法完成瓦斯浓度预测。以吉林八连城瓦斯浓度监控数据为实例,采用所提算法进行瓦斯预测,并与单一LSTM神经网络、GRU神经网络和多层感知机(MLP)进行对比。实验结果表明,对于一年(长期)南11902上顺工作面的训练集和测试集,所提算法较MLP的均方根误差(RMSE)分别降低了4.227%和3.559%;对于一年(长期)72305上顺回风的训练集和测试集,所提算法较MLP的均方根误差分别降低了7.846%和10.323%,均表现出更高的预测精度。

Aiming at the problem that statistical learning and machine learning methods were difficult to accurately predict gas concentration sequence data,a gas concentration sequence prediction algorithm based on Long Short-Term Memory(LSTM)and Gated Recurrent Unit(GRU)neural network was proposed.First,the data was divided and normalized.Then LSTM neural network cells and GRU neural network cells were introduced to process the temporal historical gas concentration sequence data.The network structure was designed to learn the internal dynamic change law of gas concentration sequence,the prediction algorithm was obtained by minimizing the error loss to complete the gas concentration prediction.Taking the monitoring data of gas concentration in Balian City of Jilin as an example,the proposed algorithm was used to predict gas,and compared with single LSTM neural network,GRU neural network and Multi-Layer Perceptron(MLP).The experimental results show that for the training set and test set of the 1-year(long-term)south 11902 upward working face,the Root Mean Square Error(RMSE)of the proposed algorithm is 4.227%and 3.559%lower than that of the MLP respectively.For the training set and test set of 1-year(long-term)72305 downwind,the RMSE of the proposed algorithm is 7.846%and 10.323%lower than that of MLP respectively,showing higher prediction accuracy.