面向富营养化评价的动态自标记半监督分类模型研究

Eutrophication is a global environmental problem and highly concerned by all over the world. It is also a key problem in the research of monitoring and early warning of ecological environment of the Three Gorges Reservoir. With the extensive applications of the water quality on-line monitoring technology, the traditional methods and models of eutrophication evaluation are facing with the challenges of high cost collecting monitoring indicators and the lack of processing capacity on real-time monitoring big data. For solving the above challenges, this project will establish a low cost, fast and accurate eutrophication evaluation model based on researching the corresponding semi-supervised classification model and technology. The main steps include: first, we will analyze the real-time monitoring big data through the density peaks to discovery the internal clustering knowledge space structure of data, and then reveal the dynamic evolution law of data; after that, we will establish a self-labeled semi-supervised classification model with the capacity of dynamically and adaptively handling real-time monitoring big data, and then use the differential evolution to optimize the model; finally, we will focus on the typical eutrophication monitoring area in Three Gorges Reservoir to develop an eutrophication evaluation prototype system which has the advantages of data driven, low cost and capacity of handling real-time monitoring big data. The results of this study will be helpful to cognize, evaluate, monitor, and early warn the eutrophication problems of the Three Gorges Reservoir.

富营养化是世界各国高度关注的一个全球性水环境问题，也是三峡库区生态环境监测预警研究中的一个关键问题。随着水质在线监测物联网技术的广泛应用，传统的富营养化评价方法与模型面临着监测指标获取代价太高和实时监测大数据处理能力不足的挑战问题。本项目针对上述问题研究相应的半监督分类模型与技术，建立低代价、快速准确的富营养化评价模型，主要包括：首先，通过密度峰值分析实时监测大数据，发现数据内部聚类知识空间结构，揭示数据的动态演化规律；进而，以数据的动态演化规律为基础，构建适应于处理实时监测大数据的动态自标记半监督分类模型，并利用差分进化优化；最后，针对三峡水库典型富营养化监测区域，研制一套数据驱动的、低代价的和适应于处理实时监测大数据的富营养化评价原型模型系统。本项目研究成果将有助于对三峡水库富营养化问题的科学认知评价和监测预警。

本项目研究聚焦于三峡库区水生态环境富营养化评价及监测预警，重点针对传统富营养化评价方法与模型面临的监测指标获取代价太高和实时监测大数据处理能力不足等挑战问题，研制一套数据驱动的、低代价的和适应于处理实时监测大数据的富营养化评价原型模型系统，包括三个方面：(1)数据聚类知识空间结构的动态演化规律研究；(2)动态自标记半监督分类模型构建及优化；(3)面向富营养化评价的动态自标记半监督分类模型构建。.本项目已经超额完成研究任务，研制了一套数据驱动的、低代价的和适应于处理实时监测大数据的富营养化评价原型模型系统；发表论文20篇，其中项目负责人以第一作者发表IEEE Trans系列汇刊4篇、中科院二区期刊1篇、CCF推荐会议3篇；申请国家发明专利2项；项目负责人获得重庆市优秀博士论文。