Scholarlink

<p>Harmful algal blooms (HABs), caused by overgrowth of algae, pose serious risks to marine ecosystems, public health, and economies. Accurate prediction of HABs is essential for effective control. Yet, due to the complex and nonlinear patterns in time series data, predicting HABs remains challenging. In response, this paper introduces the HGS-At-LSTM model, a combination of an attention-based long short-term memory (LSTM) and the halving grid search with cross-validation (HalvingGridSearchCV) optimizer. The attention mechanism helps the model focus on key temporal dependencies, while the HalvingGridSearchCV optimizes hyper-parameters. The suggested model was compared with other methods like GridSearchCV, particle swarm optimization (PSO), and RandomizedSearchCV. The experimental results show that the HGS-At-LSTM model consistently surpasses baseline approaches, achieving a coefficient of determination (<InlineEquation ID="IEq1"> <InlineMediaObject> <ImageObject Color="BlackWhite" FileRef="41060_2025_783_Article_IEq1.gif" Format="GIF" Height="16" Rendition="HTML" Resolution="72" Type="Linedraw" Width="21" /> </InlineMediaObject> <EquationSource Format="TEX">\(R^2\)</EquationSource> <EquationSource Format="MATHML"><math> <msup> <mi>R</mi> <mn>2</mn> </msup> </math></EquationSource> </InlineEquation>) value of 0.99, a mean absolute error (MAE) value between 1.12 and 2.06, and a root mean square error (RMSE) value between 2.14 and 3.11. RMSE and MAE are reduced by 26.5% and 34%, respectively, in comparison with GridSearchCV. Compared to RandomizedSearchCV, the improvements are even more pronounced, with an RMSE reduction of 43.5% and an MAE reduction of 42%. With a p-value below 0.01, the Friedman statistical test also verifies the significant efficiency of the proposed model, as it reveals that it ranks first in prediction performance. Furthermore, our model cuts training time by 40–55% compared to PSO and 65–72% compared to GridSearchCV, while still maintaining improved performance. These findings emphasize the HGS-At-LSTM model’s reliability and efficiency, making it a promising tool for accurate HAB forecasting.</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

HGS-At-LSTM: attention-based long short-term memory model combined with halving grid search optimizer for harmful algal bloom forecasting

  • 期刊论文

  • 发表日期：

    2025年5月2日

查看全文
  • Abir Loussaief,
  • Raïda Ktari,
  • Yessine Hadj Kacem

摘要

Harmful algal blooms (HABs), caused by overgrowth of algae, pose serious risks to marine ecosystems, public health, and economies. Accurate prediction of HABs is essential for effective control. Yet, due to the complex and nonlinear patterns in time series data, predicting HABs remains challenging. In response, this paper introduces the HGS-At-LSTM model, a combination of an attention-based long short-term memory (LSTM) and the halving grid search with cross-validation (HalvingGridSearchCV) optimizer. The attention mechanism helps the model focus on key temporal dependencies, while the HalvingGridSearchCV optimizes hyper-parameters. The suggested model was compared with other methods like GridSearchCV, particle swarm optimization (PSO), and RandomizedSearchCV. The experimental results show that the HGS-At-LSTM model consistently surpasses baseline approaches, achieving a coefficient of determination ( \(R^2\) R 2 ) value of 0.99, a mean absolute error (MAE) value between 1.12 and 2.06, and a root mean square error (RMSE) value between 2.14 and 3.11. RMSE and MAE are reduced by 26.5% and 34%, respectively, in comparison with GridSearchCV. Compared to RandomizedSearchCV, the improvements are even more pronounced, with an RMSE reduction of 43.5% and an MAE reduction of 42%. With a p-value below 0.01, the Friedman statistical test also verifies the significant efficiency of the proposed model, as it reveals that it ranks first in prediction performance. Furthermore, our model cuts training time by 40–55% compared to PSO and 65–72% compared to GridSearchCV, while still maintaining improved performance. These findings emphasize the HGS-At-LSTM model’s reliability and efficiency, making it a promising tool for accurate HAB forecasting.