Scholarlink

<p>Predicting total phosphorus (TP) in Poyang Lake is essential for effective water resource management and watershed ecological restoration. However, forecasting TP concentrations in Poyang Lake using machine learning faces challenges in model selection and optimization. This study assesses the impact of empirical mode decomposition (EMD) on the random forest (RF) model’s ability to evaluate hydrological and water quality factors influencing TP concentrations. Additionally, particle swarm optimization-based support vector regression (PSO-SVR) models before and after EMD processing are employed to compare four prediction models. The optimal model is integrated with an autoregressive integrated moving average (ARIMA) model to overcome machine learning limitations in capturing linear trends. Results indicate that following EMD processing, the influence of hydrological factors on TP becomes clearer, volatility decreases, and the stability of water quality factors improves. EMD enhances the RF model’s predictive accuracy, reducing MAE by 14.7%, RMSE by 6.3%, and increasing R<sup>2</sup> by 5.9%. The PSO-SVR model shows smaller improvements, and the EMD-RF model outperforms others. The EMD-RF-ARIMA model further improves TP prediction, achieving a test set MAE of 8.91 × 10<sup>− 3</sup> mg/L, RMSE of 12.20 × 10<sup>− 3</sup> mg/L, and R<sup>2</sup> of 0.836. The prediction is 64.90 × 10<sup>− 3</sup> mg/L, while the true TP is 62.50 × 10<sup>− 3</sup> mg/L, demonstrating better performance than the EMD-RF model (MAE = 8.93 × 10<sup>− 3</sup> mg/L, RMSE = 1.27 × 10<sup>− 2</sup> mg/L, R<sup>2</sup> = 0.767). Unlike previous single-model or general water quality studies, this study develops an EMD-RF-ARIMA framework for TP prediction and management support in Poyang Lake.</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

The Integrated Machine Learning Model Driven by Multi-source Data Improves the Accuracy in Lake Water Quality Prediction

  • 期刊论文

  • 发表日期：

    2026年5月28日

查看全文
  • Jiale Li,
  • Hua Wang,
  • Yanqing Deng,
  • Xiaomao Wu,
  • Yichuan Zeng

摘要

Predicting total phosphorus (TP) in Poyang Lake is essential for effective water resource management and watershed ecological restoration. However, forecasting TP concentrations in Poyang Lake using machine learning faces challenges in model selection and optimization. This study assesses the impact of empirical mode decomposition (EMD) on the random forest (RF) model’s ability to evaluate hydrological and water quality factors influencing TP concentrations. Additionally, particle swarm optimization-based support vector regression (PSO-SVR) models before and after EMD processing are employed to compare four prediction models. The optimal model is integrated with an autoregressive integrated moving average (ARIMA) model to overcome machine learning limitations in capturing linear trends. Results indicate that following EMD processing, the influence of hydrological factors on TP becomes clearer, volatility decreases, and the stability of water quality factors improves. EMD enhances the RF model’s predictive accuracy, reducing MAE by 14.7%, RMSE by 6.3%, and increasing R2 by 5.9%. The PSO-SVR model shows smaller improvements, and the EMD-RF model outperforms others. The EMD-RF-ARIMA model further improves TP prediction, achieving a test set MAE of 8.91 × 10− 3 mg/L, RMSE of 12.20 × 10− 3 mg/L, and R2 of 0.836. The prediction is 64.90 × 10− 3 mg/L, while the true TP is 62.50 × 10− 3 mg/L, demonstrating better performance than the EMD-RF model (MAE = 8.93 × 10− 3 mg/L, RMSE = 1.27 × 10− 2 mg/L, R2 = 0.767). Unlike previous single-model or general water quality studies, this study develops an EMD-RF-ARIMA framework for TP prediction and management support in Poyang Lake.