【作者】 王震；

【导师】 任南琪；

【作者基本信息】 哈尔滨工业大学 ， 环境科学与工程， 2022， 硕士

【摘要】 湿地是最富生产力的生态系统,被誉为“地球之肾”和“物种基因库”,是生物生存发展的重要基础。建设人工湿地既是保护城市湿地资源的重要途径,同时也是净化城市水系提升城市水质的有效方法。随着人工湿地建设不断加快,对人工湿地系统进行良好的预测设计是保障其净化功能的重要手段。随着机器学习方法的提出,鉴于其对数据良好的学习能力,已被用于多领域进行分类预测和回归预测。本研究根据人工湿地水质数据特点,选择了多种机器学习方法对潜流人工湿地水质进行了预测。为提升预测精度,本文基于虚拟样本对极限学习机预测模型进行了优化,完成人工湿地预测模型的建立。此外,本文构建了人工湿地生态服务价值体系,评估了生态服务价值,完成了人工湿地价值评价工作。基于上述两项工作,进行了人工湿地综合服务系统的搭建,以提升相关理论的实际应用效果。首先,基于R语言选择随机森林、Cubist、支持向量回归和极限学习机模型构建人工湿地水质预测模型,并分别对四种机器学习模型进行了参数优化。误差结果表明,极限学习机模型对人工湿地水质预测拟合误差最小,NH₃-N预测模型中RMSE为7.84,MAPE为35.68%;COD预测模型中RMSE为23.08,MAPE为18.33%。更适用于此类数据集的预测。其次,为进行水质预测模型的精度优化。将原始数据集进行类别划分,将潜流湿地划分为四种类型,并基于粒子群优化多分布扩散的虚拟样本生成技术,分别向各模型中添加虚拟样本,原始模型的精度被有效提高。随着虚拟样本个数从0增加到100个,出水NH₃-N预测模型中的RMSE平均降低了68.1%,MAPE平均降低了80.3%;出水COD预测模型中的RMSE平均降低了60.2%,MAPE平均降低了61.8%。再次,本文对长春市南溪湿地公园展开研究,核算了南溪湿地生态系统服务价值。结果表明南溪湿地生态价值总量为1.45×10⁷元/年,其中间接使用价值为1.16×10⁷元/年,占总价值比重较高为79.87%,洪水调蓄的价值最大为8.52×10⁶元/年,占总服务价值的58.59%。最后,本文基于上述工作构建了一套综合服务系统。构建了系统基本框架,对该系统进行了需求分析、总体设计,完成了以湿地水质预测和价值评估为核心的功能模块搭建及界面设计工作。更多还原

【Abstract】 Wetland is the most productive ecosystem,known as the"kidney of the earth"and"gene pool of species",is an important basis for the survival and development of organisms.The construction of constructed wetland is not only an important way to protect urban wetland resources,but also an effective method to purify urban water system and improve urban water quality.With the rapid construction of constructed wetland,a good predictive design of constructed wetland system is an important means to ensure its purification function.With the development of machine learning methods,it has been used for classification prediction and regression prediction in many fields due to its good data learning ability.In this study,according to the characteristics of water quality data of constructed wetland,a variety of machine learning methods were selected to predict the water quality of subsurface constructed wetland.In order to improve the prediction accuracy,this paper optimized the prediction model of extreme learning machine based on virtual samples and completed the establishment of the constructed wetland prediction model.In addition,this paper constructed the ecological service value system of constructed wetland,evaluated the ecological service value,and completed the value evaluation of constructed wetland.Based on the above two works,the constructed wetland comprehensive service system was built to improve the practical application effect of relevant theories.Firstly,the water quality prediction model of constructed wetland was constructed based on R language,including random forest,Cubist,support vector regression and extreme learning machine model,and the parameters of the four machine learning models were optimized respectively.The results showed that the fitting error of the extreme learning machine model for the constructed wetland water quality prediction was the smallest,and the RMSE and MAPE of NH₃-N prediction model were 7.84 and35.68%,respectively.In the COD prediction model,RMSE is 23.08 and MAPE is18.33%.More applicable to predictions from such data sets.Secondly,to optimize the precision of water quality prediction model.The original data set was classified into four types,and the sub-flow wetland was divided into four types.Virtual samples were added to each model based on the virtual sample generation technology of particle swarm optimization and multi-distribution diffusion,and the accuracy of the original model was effectively improved.With the increase of the number of virtual samples from 0 to 100,RMSE and MAPE in the effluent NH₃-N prediction model decreased by 68.1%and 80.3%on average.RMSE and MAPE in the effluent COD prediction model decreased by 60.2%and 61.8%on average.Thirdly,this paper studies Chang Chun Nanxi Wetland Park and calculates the value of ecosystem services of Nanxi wetland.The results showed that the total ecological value of Nanxi wetland was 1.45×10⁷yuan/year,of which the indirect use value was 1.16×10⁷yuan/year,accounting for 79.87%of the total value.The maximum value of flood regulation and storage was 8.52×10⁶yuan/year,accounting for 58.59%of the total service value.Finally,a set of integrated service system is constructed based on the above work.The basic framework of the system was constructed,the demand analysis and overall design of the system were carried out,and the functional module construction and interface design with wetland water quality prediction and value evaluation as the core were completed.更多还原