多层深度学习模型驱动的水质COD测量研究  

Research on Multilayer Deep Learning Model-Driven Water Quality COD Measurement

作　　者：郑培超[1] 何浩楠 陈述斌 李海娟 侯艳 李成林 阮伟 杨琴 王金梅[1] 李彪 ZHENG Peichao;HE Haonan;CHEN Shubin;LI Haijuan;HOU Yan;LI Chenglin;RUAN Wei;YANG Qin;WANG Jinmei;LI Biao(College of Optoelectronic Engineering,Chongqing University of Posts and Telecommunications,Chongqing Key Laboratory of Optoelectronic Information Sensing and Transmission Technology,Chongqing 400065,China;Chongqing Feiyang Measurement and Control Technology Research Institute Co.,Ltd.,Chongqing 400065,China)

机构地区：[1]重庆邮电大学光电工程学院,重庆市光电信息感测与传输技术重点实验室,重庆400065 [2]重庆飞扬测控技术研究院有限公司,重庆400065

出　　处：《中国无机分析化学》2025年第5期618-627,共10页Chinese Journal of Inorganic Analytical Chemistry

基　　金：国家自然科学基金资助项目(32171627);重庆市留学人员回国创业创新支持计划(启动类)项目(RC2022-56);重庆市教委科学技术研究项目(KJZD-M202200602);重庆市自然科学基金创新发展联合基金资助项目(CSTB2024NSCQ-LZX0078)。

摘　　要：有机物污染严重威胁着水资源生态系统,并直接危害人类健康。化学需氧量(COD)作为评估水体污染程度的重要指标,其准确预测对于有效的水质管理和环境保护至关重要。然而,由于水质序列的非线性和非平稳性特征,传统的预测模型在准确性上存在局限。此外,将深度学习网络与元启发式算法结合的性能尚未得到充分验证。为了克服这些挑战,提出了一种创新的多层深度学习模型,用于提高紫外-可见吸收光谱技术在COD测量中的预测精度。模型融合了卷积神经网络(CNN)以提取光谱的空间特征,双向长短期记忆网络(BiLSTM)以捕捉数据的时间依赖性,以及注意力机制(Attention)以增强对关键信息的识别。通过鲸鱼优化算法(WOA)对超参数进行优化,显著提升了预测性能。结果表明,模型在测试集上的决定系数(R2)为0.9601,均方根误差(RMSE)为0.1339,平均绝对误差(MAE)为0.1092,显著提高了COD预测的精确度和鲁棒性。未来的研究将探索集成更多环境变量,以开发更为全面的深度学习模型,进一步推动水质监测与管理技术的发展。Organic pollution poses a severe threat to the ecosystem of water resources and directly endangers human health.Chemical Oxygen Demand(COD),as an important indicator for assessing the degree of water body pollution,its accurate prediction is crucial for effective water quality management and environmental protection.However,traditional predictive models have limitations in accuracy due to the nonlinear and non-stationary characteristics of water quality sequences.Moreover,the performance of combining deep learning networks with metaheuristic algorithms has not been fully verified.To overcome these challenges,this paper proposes an innovative multi-layer deep learning model to improve the predictive accuracy of COD measurements using ultraviolet-visible absorption spectroscopy.The model integrates Convolutional Neural Networks(CNN) to extract spatial features of the spectrum,Bi-directional Long Short-Term Memory Networks(BiLSTM) to capture the temporal dependencies of the data,and an Attention mechanism to enhance the recognition of key information.By optimizing hyperparameters with the Whale Optimization Algorithm(WOA),the predictive performance is significantly enhanced.The results showed that the model achieved a coefficient of determination(R2) of 0.960 1,a root mean square error(RMSE) of 0.133 9,and a mean absolute error(MAE) of 0.109 2 on the test set,significantly improving the precision and robustness of COD prediction.Future research will explore the integration of more environmental variables to develop a more comprehensive deep learning model,further promoting the development of water quality monitoring and management technology.

关 键 词：光谱学 化学需氧量 深度学习 超参数优化 水质检测 

分 类 号：O433.4[机械工程—光学工程]