【摘要】 为提高瓦斯涌出量预测精度，提出一种基于变分模态分解(variationalmode decomposition,VMD)的注意力机制(attentionmechanism,AM)-长短期记忆(longshortterm memory, LSTM)网络与极端梯度提升(extreme gradient boosting, XGBoost)组合的预测模型。利用VMD将瓦斯涌出量原始数据分解为高、低频率的分量，以长短期记忆网络时序分析模型为基础，将分解后的高频分量作为其输入。同时，引入注意力机制提取瓦斯涌出量影响因素时序数据中的关键信息，增强序列数据中关键信息的表达，提高模型的预测精度。利用XGBoost模型对低频分量进行预测，将高、低频分量的预测结果进行叠加求和，得到最终的瓦斯涌出量预测值。根据实验结果，引入注意力机制后模型的预测精度明显高于无注意力机制的预测模型，且所提出的组合模型的预测精度均高于对应的单一模型和其他对比模型，验证了该方法的有效性。更多还原

【Abstract】 In order to improve the prediction accuracy of gas emission, a combined AM-LSTM and XGBoost prediction model based on variational modal decomposition(VMD) is proposed. The VMD is utilized for decomposing the original data of gas emission into high and low frequency components. The decomposed high-frequency components are used as input to the model based on LSTM timing analysis. At the same time,on the basis of the Long Short Term Memory, an Attention Mechanism that can autonomously extract the key information in the gas emission time series data is introduced to enhance the expression of the key information in the sequence data and improve the prediction accuracy of the model. The XGBoost model is built for the purpose of predicting the low-frequency components, and then the prediction results of the high and low-frequency components are efficiently overlaid to obtain the final gas emission prediction value. In accordance to the experimental results, the prediction accuracy of the model after the introduction of the Attention Mechanism is higher, which is significantly higher than that of the prediction model without the Attention Mechanism. Moreover, compared with the single prediction model and other intelligent algorithm models, the prediction accuracy of the proposed combined model is significantly higher, which verifies the effectiveness of the method.更多还原