摘要
准确的水质预报及关键因素识别是流域水环境控制及水资源保护的重要依据。针对地表水水质评价指标的多样性以及原始数据的非线性、非平稳的特征,在改善传统的分解-预测-集成时间序列预测框架之上,提出了一种基于集合经验模态分解(EEMD)和灰狼优化集成策略(GWO)数据预测模型。该模型采用集合经验模态分解方法对数据进行分解,得到不同频率的分解分量,再分别运用Elman神经网络、长短记忆神经网络(LSTM)和支持向量回归(SVR)作为基学习器进行预测,然后利用灰狼动态优化集成策略对预测结果进行集成,最后,以长江下游水质断面的数据对模型有效性进行评估,并与单一预测模型和平均集成策略的预测模型进行对比。研究结果表明,该模型在误差评价指标和DM检验上表现优异,相较于其他模型更具优势,其中,对于溶解氧的预测,相较于平均集成策略的预测模型,通过灰狼优化集成模型的平均绝对误差(MAE)、均方根误差(RMSE)以及Theil不等系数(TIC)分别降低了16.29%、13.17%、13.24%,对于最终水质等级的预测正确率为98.6%,证明该模型能够准确预测水质等级并识别关键影响因素,进而为水环境污染治理提供科学依据。
Accurate water quality prediction and identification of key factors are important foundations for water environment control and water resource protection in river basins.Considering the diversity of surface water quality evaluation indicators and the nonlinear and non-stationary characteristics of raw data,an improved ensemble empirical mode decomposition(EEMD)and grey wolf optimization(GWO)integrated data prediction model is proposed based on the traditional decompose-forecast-integrate time series prediction framework.This model utilizes the ensemble empirical mode decomposition method to decompose the data into different frequency components.Then,Elman neural network,long short-term memory neural network(LSTM),and support vector regression(SVR)are employed as base learners for individual predictions.Subsequently,a grey wolf dynamic optimization integration strategy is applied to integrate the prediction results.Finally,the model′s effectiveness is evaluated using data from a water quality section in the lower reaches of the Yangtze River,and compared with single prediction models and average integration strategies.The research results show that the proposed model performs excellently in error evaluation metrics and DM test.It outperforms other models,especially in the prediction of dissolved oxygen.Compared to the average integration strategy prediction model,the grey wolf optimization integration model reduces the mean absolute error(MAE),root mean square error(RMSE),and Theil inequality coefficient(TIC)by 16.29%,13.17%,and 13.24%respectively.The model achieves a prediction accuracy of 98.6%for final water quality grades,indicating its capability to accurately predict water quality levels and identify key influencing factors,thus providing a scientific basis for water environment pollution control.
作者
李明
赵良伟
蒋一波
刘东岳
LI Ming;ZHAO Liang-wei;JIANG Yi-bo;LIU Dong-yue(Business School of Hohai University,Nanjing 211100,Jiangsu Province,China;Institute of Project Management Informatization,HoHai University,Nanjing 211100,Jiangsu Province,China;Jiangsu Huaiyin Water Conservancy Construction Co.,LTD,Jiangyin 223005,Jiangsu Province,China)
出处
《中国农村水利水电》
北大核心
2024年第11期95-102,共8页
China Rural Water and Hydropower
基金
国家社会科学基金项目“工程建设市场主体社会化监管机制研究”(17BGL156)
河海大学中央高校基本科研业务费项目“战略视角下长江大保护水环境基础设施治理模式及对策研究”(B220207039)。
关键词
水质预测
集合经验模态分解
灰狼优化
长江下游
集成策略
water quality predictions
EEMD
grey wolf optimizer
the lower reaches of the Yangtze River
integration strategy
