基于多头注意力机制的瓦斯多粒度预测方法

A Gas Multi-Granularity Prediction Method Based on Multi-Head Attention Mechanism

瓦斯是影响矿井安全的重要因素,但现有瓦斯预测工作忽略多粒度数据的异质性,使得预测精度不高。单粒度数据不能完全表示出瓦斯变化的特征,且现有方法不能完全挖掘不同粒度下的数据特性。基于多粒度思想,通过CNN聚合构建多粒度数据,并借助LSTM与多头自注意力的特征提取能力,提出了基于多头自注意力机制的瓦斯多粒度预测模型(MGPM)。上述模型能够有效满足瓦斯预测任务中对不同粒度数据的构建,实现煤矿瓦斯数据在不同粒度特性下的深入挖掘。实验结果表明,所提出的模型相比与基线模型降低了预测误差。

Gas is an important factor affecting mine safety,but the existing gas prediction work ignores the heterogeneity of multi-granularity data,resulting in low prediction accuracy.Single granularity data cannot fully express the characteristics of gas change,and the existing methods cannot fully mine the data characteristics under different granularity.Based on the idea of multi-granularity,multi-granularity data is constructed by CNN aggregation,and with the feature extraction ability of LSTM and multi-head self-attention mechanism,a coal mine gas multi-granularity prediction model based on multi-head self-attention mechanism(MCPM)is proposed.The model can effectively meet the construction of different granularity data in gas prediction tasks,and realize the in-depth mining of coal mine gas data under different granularity characteristics.The experimental results show that the proposed model reduces the prediction error compared with the baseline model.