|
| Predicting Acid Concentration of In-situ Leaching Uranium Solutions Based on iAutoformer |
|
Received:October 13, 2024
Revised:November 13, 2024
Accepted:November 14, 2024
Published Online:April 19, 2025
|
| View Full Text View/Add Comment Download reader |
| DOI：doi：10.20237/j.issn.1007-7545.2025.05.011 |
| KeyWord:in-situ leaching uranium；deep learning；auto-correlation mechanism；time series forecasting；Transformer |
| Author | Institution |
| LIU Zhifeng |
东华理工大学 核资源与环境国家重点实验室 |
| LIN Zhining |
东华理工大学 信息工程学院 |
| YANG Yihan |
中核内蒙古矿业有限公司 |
| WANG Ruyi |
中核内蒙古矿业有限公司 |
| ZHANG Chuanfei |
中核内蒙古矿业有限公司 |
| LIANG Daye |
中核内蒙古矿业有限公司 |
| TANG Junxian |
东华理工大学 信息工程学院 |
|
| Hits: 519 |
| Download times: 1340 |
| Abstract: |
| Under the goal of carbon peaking and carbon neutrality, the efficiency of uranium mining will directly affect the development of nuclear energy and the progress of national energy transformation. In in-situ acid leaching of uranium, accurately predicting the acid concentration of leaching solution has practical significance for improving leaching rate of uranium, protecting the environment of uranium ore and leaching equipment. In this paper, iAutoformer model, a framework based on the auto-correlation mechanism and self-attention mechanism, was proposed. By weighting and rolling aggregation of historical time series autocorrelation, the periodicity of the sequence was discovered and more hidden factors were revealed, the utilization of information was expanded, and better prediction results were obtained. The experimental average results show that the iAutoformer model can discover sequence variation patterns in small numerical fluctuations compared to the Long Short Term Memory (LSTM, 2000) model, the Mean Absolute Error (MAE) drops from 0.232 to 0.205, and the Mean Square Error (MSE) drops from 0.089 to 0.072. Compared to the Inverted Transformer (iTransformer, 2023) and Patch Time Series Transformer (PatchTST, 2022) models, the iAutoformer model fits real values more accurately in predictions with significant periodic changes, with MAE decreasing by 0.005 and 0.004, and MSE both decreasing by 0.003, respectively. A prediction method is proposed from the perspective of the production environment of in-situ uranium leaching. By learning the patterns from historical data to make real-time predictions of future leachate acid concentrations, it can effectively optimize and adjust the acid concentration, enhance production efficiency, and facilitate the intelligent transformation of uranium mines. |
| Close |
|
|
|