基于GA-SVM的渤海湾富营养化模型  

Eutrophication Model of Bohai Bay Based on GA-SVM

作　　者：向先全[1] 陶建华[2] 

机构地区：[1]天津大学环境科学与工程学院,天津300072 [2]天津大学机械工程学院,天津300072

出　　处：《天津大学学报》2011年第3期215-220,共6页Journal of Tianjin University(Science and Technology)

基　　金：国家自然科学基金资助项目(10872144)

摘　　要：为了更好地模拟和认知渤海湾富营养化的复杂行为,通过研究遗传算法(GA)和支持向量机(SVM)的结合形式,即参数寻优和特征选择,以渤海湾水质实测资料为依据,叶绿素a的质量浓度为输出,建立了GA-SVM的富营养化模型.无特征选择时,用遗传算法对支持向量机的参数(惩罚参数和核参数)进行自适应地优选,预测模型的均方误差可达到1.831μg/L,具有较好的认知、泛化能力.再利用遗传算法二进制编码及启发式寻优的优点,对所建模型的输入空间进行特征选择,提取出代表性的特征变量:DO%、pH值、水温、COD、盐度以及氨氮.特征提取后预测模型的均方误差可达到1.363μg/L,模型性能有了很大提高.分析表明,COD、盐度及氨氮可作为人为控制的首要指标.For better simulating and cognizing the complex eutrophication behaviors of Bohai Bay,the combining forms of genetic algorithm and support vector machine（parameter optimization and feature selection） have been re-searched to establish GA-SVM eutrophication model for Bohai Bay based on the field measured data,and chloro-phyll_a content has been selected as the model output.Firstly,with GA self-adaptive optimizing for penalty parameter and kernel parameter,the root mean square error（RMSE） of SVM test model was 1.831 μg/L,indicating preferable generalization performance.Then,with GA-based feature selection for the established SVM model,RMSE of SVM test model was 1.363 μg/L,showing great improvement for model performance.The representative features were ex-tracted such as DO%,pH,water temperature,COD,salinity,and ammonia-nitrogen,the latter three of which could be considered as prior indexes for artificial control of eutrophication based on further analysis.

关 键 词：富营养化模型 支持向量机 遗传算法 参数寻优 特征选择 渤海湾 

分 类 号：X171[环境科学与工程—环境科学] TP181[自动化与计算机技术—控制理论与控制工程]