Purpose

Dust pollution is currently one of the most serious environmental problems faced by open-pit mines. Compared with underground mining, open-pit mining has many dust sources, and a wide area of influence and complicated changes in meteorological conditions can result in great variations in dust concentration. Therefore, the prediction of dust concentrations in open-pit mines requires research and is of great significance for reducing environmental pollution and personal health hazards.

Methods

This study is based on monitoring of the concentration of total suspended particulate (TSP) in the Anjialing open-pit coal mine in Pingshuo. This paper proposes a hybrid model based on a long short-term memory (LSTM) network and the attention mechanism (LSTM-Attention) and applies it to the prediction of TSP concentration. The LSTM model reflects the historical process of an input time series, and the attention mechanism extracts the inherent characteristics of the input parameters to assign weights based on the importance of the influencing factors. The autoregressive integrated moving average (ARIMA) and LSTM models are also used to predict the TSP concentration. Finally, several statistical measures of error are used to evaluate the accuracy of the model and perform a sensitivity analysis.

Results

It was found that, in general, the TSP concentration was highest in the period 08:00–09:00 and lowest in the period 15:00–16:00. In addition to the influence of meteorological parameters and normal operations, the reason for this trend is the presence of an inversion layer above the open-pit mine. The results show that, compared with the ARIMA and LSTM models, the LSTM-Attention model is more stable and has a prediction accuracy that is 5.6% and 3.0% greater, respectively.

Conclusion

This model can be applied to the prediction of dust concentrations in open-pit mines and provide guidance on when to carry out dust-suppression work. It has expansibility and is potentially valuable for application in a wide range of areas.

中文翻译：

---

基于深度学习算法的露天矿粉尘浓度监测与预测

目的

粉尘污染是目前露天矿山面临的最严重的环境问题之一。与地下开采相比，露天开采粉尘源多，影响范围广，气象条件变化复杂，导致粉尘浓度变化较大。因此，露天矿粉尘浓度的预测需要研究，对于减少环境污染和人身健康危害具有重要意义。

方法

本研究基于对平朔安家岭露天煤矿总悬浮颗粒物（TSP）浓度的监测。本文提出了一种基于长短期记忆（LSTM）网络和注意力机制（LSTM-Attention）的混合模型，并将其应用于TSP浓度的预测。LSTM模型反映了输入时间序列的历史过程，注意力机制提取输入参数的固有特征，根据影响因素的重要性分配权重。自回归综合移动平均 (ARIMA) 和 LSTM 模型也用于预测 TSP 浓度。最后，使用几个统计误差度量来评估模型的准确性并进行敏感性分析。

结果

结果发现，总体而言，TSP 浓度在 08:00-09:00 期间最高，在 15:00-16:00 期间最低。除了气象参数和正常作业的影响外，造成这种趋势的原因是露天矿上方存在逆温层。结果表明，与 ARIMA 和 LSTM 模型相比，LSTM-Attention 模型更稳定，预测准确率分别提高了 5.6% 和 3.0%。

结论

该模型可应用于露天矿山粉尘浓度预测，为何时开展抑尘工作提供指导。它具有可扩展性，在广泛的领域中具有潜在的应用价值。