基于多源信息融合的水质在线异常检测与分类识别方法研究

The online detection, classification and identification of water-quality contamination events during sudden water pollution accidents are critical to improve the performance for in-situ water-quality monitoring systems and early-warning systems. This project presents a novel approach for detecting and classifying anomaly water quality events based on multi-source information fusion. Base on the correlation analysis of the measured multi-source water quality indicators and the contaminants, three key scientific problems, including online water-quality anomaly detection based on common water-quality parameters and spectral data, online water-quality anomaly detection in water distribution system, and the classification and identification of water anomalies and contaminants, are researched by introducing spectral data, adjacent node information, and fingerprint features. A series of water-quality anomaly detection models and classification models are studied by optimizing selections. Based on the technical achievements of the project, a set of online early-warning approaches and solutions are developed for the sudden water pollution accidents. The research results of the project would lay a solid theoretical and application foundation for the security of urban water supply in new era.

突发水质污染事件下的水质在线异常检测和分类识别技术是提升现有水质在线监测系统预警能力的关键。本项目拟开展基于多源信息融合的水质在线异常检测和分类识别方法研究。以突发污染事件下多源水质指标与污染物的相关性为基础，充分利用光谱数据、相邻关联站点、污染指纹特征等信息，研究常规水质与光谱数据联用的水质在线异常检测、分布式传感网络下的水质在线异常检测、突发污染事件下的水质异常分类和识别等关键问题。探索在这些问题下的优化异常检测和分类识别模型，形成基于多源信息融合的突发水质污染事件智能在线预警新手段、新方法和新方案。为新时期城镇供水安全保障工作提供有力的理论支撑和应用指导。

近年来，随着我国经济的快速发展，水污染问题日益严重，突发性水污染事件频繁发生，直接影响饮用水安全。由于水环境的复杂性、污染物的多样性和突发事件的随机性，现有大多数水质监测预警系统根据单一水质指标进行超标阈值报警的现状，难以满足进行水质污染事件智能检测和自动检测的需要。本项目以水质在线分析仪器获得的水质指标数据为主要源信息，从拓展指标维度和空间维度、提高异常检测性能角度出发，开展基于多源信息融合的水质在线异常检测和分类识别方法研究。以突发污染事件下多源水质指标与污染物的相关性为基础，充分利用光谱数据、相邻关联站点、污染指纹特征等信息，解决基于常规水质和光谱数据的水质在线异常检测、分布式传感网络下的水质在线异常检测、突发污染事件下的水质异常分类和识别等关键问题，形成面向不同应用场景的水质在线异常检测和分类识别模型，研发完成基于多源信息融合的智能水质预警原型系统，为突发事件水质污染事件提供快速、可靠的在线预警新手段、新方法和新方案，为供水安全保障工作提供有力支撑。项目提出了一系列面向不同应用场景的水质异常预警检测算法，自主开发完成水质异常检测算法库及应用软件，项目执行期间共发表学术论文20篇，其中SCI收录期刊论文13篇，国内核心期刊论文1篇，国内学术会议论文1篇，国际学术会议论文5篇（4篇EI收录），申请国家发明专利8项（已授权1项）。