| 基于深度学习的地下水水质预测方法研究——以大湾区(广州)地下水多层监测基地氨氮为例 |
| Research on Groundwater Quality Prediction Method Based on Deep Learning:The Case of Multi-layer Groundwater Monitoring Base,Guangzhou,Greater Bay Area |
| 投稿时间：2024-02-06 修订日期：2024-04-23 |
| DOI：10.19316/j.issn.1002-6002.2025.01.20 |
| 中文关键词: 地下水水质预测 集成经验模态分解 深度学习 |
| 英文关键词:groundwater quality prediction ensemble empirical mode decomposition deep learning |
| 基金项目:国家重点研发计划项目(2022YFC3700905) |
|
| 通讯作者:李名升* 中国环境监测总站, 国家环境保护环境监测质量控制重点实验室, 北京 100012 |
| 摘要点击次数: 197 |
| 全文下载次数: 60 |
| 中文摘要: |
| 地下水水质预测可以准确反映水质的未来变化趋势,是地下水污染防治中的重要环节。为提高地下水水质自动监测的预测精度,以大湾区(广州)地下水多层监测基地2022年3层地下水的氨氮监测结果为例,进行数据预处理后,构建将集成经验模态分解(EEMD)与深度学习技术(RNN、LSTM及GRU)相结合的复合深度学习模型,开展传统机器学习模型与不同深度学习模型在水质预测效果方面的对比分析,并探讨深度学习模型多步长预测的预测效果。研究结果表明:①与单一深度学习模型相比,结合EEMD的复合深度学习模型解决了预测滞后性问题,具有更高的预测精度和拟合度。②复合深度学习模型的拟合度高于4种传统机器学习模型。4种传统机器学习模型中,仅MLR与RF的预测拟合度与单一深度学习模型接近。③多步长预测结果表明,复合深度学习模型可以准确地预测地下水水质在3 d内的变化趋势。综上,复合深度学习模型展现出更好的预测性能和泛化能力,可为地下水水质预测提供支撑。 |
| 英文摘要: |
| Groundwater quality prediction can reflect the future trend of groundwater quality changes,and is an important link in groundwater pollution prevention and control. To improve the accuracy of water quality prediction in automatic groundwater monitoring,the ammonia nitrogen levels of three layers of groundwater monitored at the multi-layer monitoring base in the Greater Bay Area (Guangzhou) from March to December 2022 were selected, a composite deep learning model combining integrated empirical mode decomposition (EEMD) and deep learning techniques (RNN, LSTM, and GRU) was constructed after data preprocessing. The effect of traditional machine learning model and different deep learning models in water quality prediction were compared and analyzed, and the prediction effect of multi-step prediction of deep learning model was discussed. The results showed that:① Compared with a single deep learning model,the composite deep learning model solves the problem of prediction lag and has higher prediction accuracy and fitness. ② Fitness of the composite deep learning model is higher than that of the four traditional machine learning models. In the four traditional machine learning models, only the fitness of MLR and RF predictions is close to that of the single deep learning prediction. ③ The multi-step prediction results indicated that the composite deep learning model can accurately predict the change trend of groundwater quality within 3 days. The composite deep learning model exhibits better predictive performance and generalization ability,which can provide support for groundwater quality prediction. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
