Abstract:

Due to the traditional prediction method of gas concentration in working face using only temporal characteristics of gas data, lacking of prior information related to space, in this paper, using the space-time characteristics of gas data, the prediction model of LSTM-FC（Long Short Time Memory-Full Connection） gas concentration time-space series is constructed by combining long short time memory with full connection neural network. LSTM can solve the problem of long-term dependence gas sequence. Fully connection neural network can accurately capture the spatial correlation of gas sequences. The combination of the two can deeply dig the space-time characteristics of gas data. Then the gas distribution diagram of the working face is constructd by predicting the gas values at different locations. The experimental results show that, by using LSTM-FC model, the prediction error is significantly reduced, and the prediction accuracy is improved compared with other neural network prediction models.

Key words: LSTM-FC, spatiotemporal sequences, gas concentration, neural network

摘要：

传统的工作面瓦斯预测方法仅利用瓦斯数据的时间特性，缺乏与空间相关的先验信息，因此利用瓦斯数据的时空特性，采用深度学习算法长短期记忆与全连接神经网络相结合的方法构建LSTM-FC（Long Short Time Memory-Fully Connection）瓦斯浓度时空序列的预测模型。LSTM能够解决瓦斯序列的长时间依赖性，全连接神经网络能够准确捕捉瓦斯序列的空间关联性，深入挖掘瓦斯数据之间的时空特性，通过预测不同位置的瓦斯值，构造工作面的瓦斯分布图。实验结果表明，通过使用LSTM-FC模型，预测误差有了明显减少，相比于其他神经网络预测模型，预测精度有所提高。

关键词: LSTM-FC, 时空序列, 瓦斯浓度, 神经网络