摘要
科学有效地评估地表水的水质对于水资源管理和人类健康具有重要意义。提出了一种基于改进哈里斯鹰优化算法(Harris Hawk Optimization,HHO)优化LightGBM,并结合卷积神经网络(Convolutional Neural Network,CNN)与LSTM(Long Short-Term Memory,LSTM)的水质分类方法。利用改进HHO优化LightGBM超参数,提升其计算效率与分类性能;同时构建CNN-LSTM模型以捕捉水质数据中的深层特征关联。为充分利用不同模型的优势,采用堆叠(Stacking)策略,将CNN-LSTM与优化后的LightGBM作为基学习器进行融合。实验结果表明:集成模型在分类准确率、召回率和F1分数等指标上,较单一模型平均提升2.7%、3.6%和3.2%。在处理复杂水质特征方面表现优异,分类准确性更高。对水质分类研究具有参考价值,有助于提高水质管理水平与决策效率。
Scientifically and effectively evaluating surface water quality is of great significance for water resource management and human health.This study proposes a water quality classification method based on an improved Harris Hawk Optimization(HHO)algorithm to optimize LightGBM and Long Short-Term Memory(LSTM)networks.The improved HHO algorithm is used to optimize the hyperparameters of LightGBM,enhancing both computational efficiency and classification performance.Additionally,Convolutional Neural Networks(CNN)combined with LSTM are employed to extract deep feature correlations from water quality data.To fully utilize the advantages of different models,a Stacking strategy is adopted,integrating CNN-LSTM and the optimized LightGBM as base learners for more efficient water quality classification.Experimental results demonstrate that the ensemble model achieves average improvements of 2.7%,3.6%,and 3.2%in classification accuracy,recall,and F1-score,respectively,compared to single models(CNN-LSTM and optimized LightGBM).The ensemble model performs well in handling complex water quality features,showing superior classification accuracy.This approach provides valuable insights into water quality classification research and contributes to improving water resource management and decision-making efficiency.
作者
罗缘
朱文忠
吴宇浩
LUO Yuan;ZHU Wenzhong;WU Yuhao(School of Computer Science and Engineering,Sichuan University of Science&Engineering,Yibin Sichuan 643002,China)
出处
《兰州工业学院学报》
2025年第6期99-105,共7页
Journal of Lanzhou Institute of Technology
基金
四川省科技计划重点研发项目(2023YFS0371)
企业信息化与物联网测控技术四川省高校重点实验室开放基金(2024WYJ03)
四川省智慧旅游研究基地(ZHYJ24-01)。
