Scholarlink

<p>Precise prediction of gas concentration holds immense importance in averting and managing calamitous coal mine gas incidents. However, achieving long-term predictions of gas concentration in the working face proves to be a challenge in the complex geological environment of deep coal mining. This study presents a multi-source information fusion prediction method for gas concentration in the mining face based on Informer. The approach relies on the SPCA algorithm to create a multi-integration index system that combines multiple factors (such as coal seam gas content, drilling cuttings weight, wind speed, and gas sensors) with a strong correlation to gas concentration in the working face. Furthermore, it establishes an Informer prediction model based on this multi-indicator system, enabling multi-step forecasting of gas concentration in the working face. The findings demonstrate that the model is capable of predicting gas concentration. A comparison between the predictions generated by the multi-integration index system established in this paper and those solely based on sensor data indicates that the former significantly enhances the accuracy of the predictions. The mean absolute error loss (MAE) per step for the Informer model does not vary by more than 0.001. Compared with the prediction obtained by using the traditional index system, the mean absolute error loss of the fifth step of the prediction results obtained by using the multi-integrated index system is reduced from 0.0248 to 0.0092. The model is also capable of multi-step prediction, with a maximum achievable prediction length of ten steps. The mean absolute error of the predictions remains consistently below 0.011, signifying an excellent fit. A comparison with the predictions from the LSTM and CNN models demonstrates that the Informer model proposed in this paper achieves a maximum mean absolute error of 0.0106, which is notably lower than the maximum mean absolute errors of 0.0346 for the LSTM model and 0.0455 for the CNN model. These results underscore the model’s efficacy in accurately forecasting fluctuations in gas concentration at the working face. The research findings furnish a significant theoretical foundation for predicting gas disasters.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-information Fusion Gas Concentration Prediction of Working Face Based on Informer

  • 期刊论文

  • 发表日期：

    2025年2月5日

查看全文
  • Binglong Liu,
  • Zhonghui Li,
  • Zesheng Zang,
  • Shan Yin,
  • Yue Niu,
  • Minbo Cai

摘要

Precise prediction of gas concentration holds immense importance in averting and managing calamitous coal mine gas incidents. However, achieving long-term predictions of gas concentration in the working face proves to be a challenge in the complex geological environment of deep coal mining. This study presents a multi-source information fusion prediction method for gas concentration in the mining face based on Informer. The approach relies on the SPCA algorithm to create a multi-integration index system that combines multiple factors (such as coal seam gas content, drilling cuttings weight, wind speed, and gas sensors) with a strong correlation to gas concentration in the working face. Furthermore, it establishes an Informer prediction model based on this multi-indicator system, enabling multi-step forecasting of gas concentration in the working face. The findings demonstrate that the model is capable of predicting gas concentration. A comparison between the predictions generated by the multi-integration index system established in this paper and those solely based on sensor data indicates that the former significantly enhances the accuracy of the predictions. The mean absolute error loss (MAE) per step for the Informer model does not vary by more than 0.001. Compared with the prediction obtained by using the traditional index system, the mean absolute error loss of the fifth step of the prediction results obtained by using the multi-integrated index system is reduced from 0.0248 to 0.0092. The model is also capable of multi-step prediction, with a maximum achievable prediction length of ten steps. The mean absolute error of the predictions remains consistently below 0.011, signifying an excellent fit. A comparison with the predictions from the LSTM and CNN models demonstrates that the Informer model proposed in this paper achieves a maximum mean absolute error of 0.0106, which is notably lower than the maximum mean absolute errors of 0.0346 for the LSTM model and 0.0455 for the CNN model. These results underscore the model’s efficacy in accurately forecasting fluctuations in gas concentration at the working face. The research findings furnish a significant theoretical foundation for predicting gas disasters.