Scholarlink

<p>To improve the accuracy of uranium concentration prediction and the scientific regulation of acid injection concentration in In-situ leaching uranium mining, this study proposes a deep learning model that integrates Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and an attention mechanism (CNN–GRU–Attention). The proposed model leverages selected key features as inputs, integrating the local feature extraction capabilities of CNNs, the temporal sequence modeling strengths of GRUs, and the critical information enhancement provided by the attention mechanism to dynamically forecast trends in extraction uranium concentration. Experimental results show that the proposed model achieves a mean absolute error (MAE) of 0.0543, a mean squared error (MSE) of 0.0315, and a coefficient of determination (<InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(\hbox {R}^{2}\)</EquationSource> </InlineEquation>) of 0.915 on the test set, outperforming LSTM, CNN–LSTM, and Transformer models, particularly in sections with abrupt concentration changes. Ablation experiments further confirm the significant contributions of the CNN and attention modules to model performance. Single-variable simulation analysis based on the model reveals that uranium concentration exhibits a “rise-then-stabilize” pattern with increasing acid injection concentration. The optimal acid injection concentrations under different residual uranium reserves are determined as 11.69 g/L, 10.85 g/L, 11.41 g/L, and 10.99 g/L, respectively, achieving production optimization while balancing leaching efficiency and cost.</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

Prediction of extraction uranium concentration and optimization of injection acid concentration using a CNN–GRU–attention model in in-situ uranium leaching

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月24日

查看全文
  • Zhenhua Wei,
  • Long Wang,
  • Zhifeng Liu,
  • Yipeng Zhou

摘要

To improve the accuracy of uranium concentration prediction and the scientific regulation of acid injection concentration in In-situ leaching uranium mining, this study proposes a deep learning model that integrates Convolutional Neural Networks (CNN), Gated Recurrent Units (GRU), and an attention mechanism (CNN–GRU–Attention). The proposed model leverages selected key features as inputs, integrating the local feature extraction capabilities of CNNs, the temporal sequence modeling strengths of GRUs, and the critical information enhancement provided by the attention mechanism to dynamically forecast trends in extraction uranium concentration. Experimental results show that the proposed model achieves a mean absolute error (MAE) of 0.0543, a mean squared error (MSE) of 0.0315, and a coefficient of determination ( \(\hbox {R}^{2}\) ) of 0.915 on the test set, outperforming LSTM, CNN–LSTM, and Transformer models, particularly in sections with abrupt concentration changes. Ablation experiments further confirm the significant contributions of the CNN and attention modules to model performance. Single-variable simulation analysis based on the model reveals that uranium concentration exhibits a “rise-then-stabilize” pattern with increasing acid injection concentration. The optimal acid injection concentrations under different residual uranium reserves are determined as 11.69 g/L, 10.85 g/L, 11.41 g/L, and 10.99 g/L, respectively, achieving production optimization while balancing leaching efficiency and cost.