基于深度学习的漂浮水稻种植区水质参数预测

Prediction of Water Quality Parameters in Floating Rice Cultivation Areas Based on Deep Learning

摘要: 目的：采煤沉陷区是当前生态修复与治理的典型难题，沉陷区环境承载力脆弱，极易受到污染。为探究“漂浮水稻”种植对采煤沉陷积水区的影响，同时提升对采煤沉陷积水区的监测能力，研究以淮南市顾桥矿采煤沉陷积水区“漂浮水稻”种植试验区为研究对象，对种植了“漂浮水稻”的被修复水域进行水质监测和分析，为采煤沉陷积水区水体的监测提供技术支持。方法：研究通过收集水体的总磷、总氮、氨氮和叶绿素a等指标，分析并评价了种植区水域的水质情况，同时构建了门控循环单元(GRU)模型。通过优化模型的特征选择和参数调整，提升预测精度和泛化能力，实现对采煤沉陷积水区水体指标的时间序列预测。结果：研究表明，利用“漂浮水稻”能一定程度上净化采煤沉陷积水区水体，遏制水体富营养化。研究结果显示，GRU模型对叶绿素a的预测效果最优，决定系数R²为0.89，与对总磷、总氮和氨氮的预测相比，GRU模型对叶绿素a的预测具备更高的可信度。结论：研究成功构建了对采煤沉陷积水区水体指标的时间序列预测模型，为采煤沉陷积水区水环境监测提供了技术支持。

Abstract: Objective: Coal mining subsidence areas are typical challenges in current ecological restoration and management. The environmental carrying capacity of subsidence areas is fragile and highly susceptible to pollution. To explore the impact of “floating rice” cultivation on waterlogged areas in coal mining subsidence zones, and to enhance the monitoring capability of such areas, this study focused on the floating rice cultivation experimental zone in the Guqiao Mine waterlogged area of Huainan City. Water quality monitoring and analysis were conducted in the restored waters where “floating rice” was planted, providing technical support for the monitoring of water bodies in coal mining subsidence areas. Methods: The study collected indicators such as total phosphorus, total nitrogen, ammonia nitrogen, and chlorophyll a to analyse and evaluate the water quality of the cultivation area. In addition, a Gated Recurrent Unit (GRU) model was constructed. By optimising feature selection and adjusting parameters, the model’s prediction accuracy and generalisation capability were enhanced, enabling time-series prediction of water indicators in coal mining subsidence waterlogged areas. Results: The study indicated that the use of “floating rice” can purify water in coal mining subsidence areas to a certain extent and curb eutrophication. The results showed that the GRU model had the best predictive performance for chlorophyll a with a coefficient of determination R² of 0.89. Compared with the predictions for total phosphorus, total nitrogen, and ammonia nitrogen, the GRU model provided higher reliability for chlorophyll a prediction. Conclusion: The study successfully established a time-series prediction model for water indicators in coal mining subsidence waterlogged areas, offering technical support for water environment monitoring in such zones.