基于序列挖掘与智能计算的地下水突发性污染源的发现与反演

The burying characteristic of underground water and the dispersion of emergent contaminant sources in space and time make it difficult to discover and inverse the contaminant source in time. Firstly, the study will adopt modern spatio-temporal sequence mining methods to establish the comprehensive model database for water quality sequence. This database will have fine description ability and complete the mapping from the space of water quality parameters to the space of sequence models based on the high dimensional spatio-temporal database. Secondly, this study will establish the sequence model set adaptive strategy of water quality and multi-model cooperative working mechanism under the guidance of the nonlinear system estimation theory, at the same time, complete the multi-dimensional prediction of groundwater parameters and the discovery of water quality abnormity; Finally, this study will convert the problem of groundwater contaminant inversion into optimization problem whose decision variables are pollution source location and strength. Then, the synchronization inversion framework including simulation models and optimization inversion models will be established. The improved intelligent search methods will be produced to get the position and the type of the groundwater contaminant, through the fast search for optimal solution in the global range. The research will provide theoretical basis and a new solution for the discovery, location and identification of groundwater contaminant source.

地下水环境的掩埋性和突发性污染源的时-空分散性，使地下水突发性污染源的发现和反演一直尚未得到解决。本课题以地下水高维时-空数据库为基础，采用现代时-空序列挖掘方法，建立具有精细描述能力的水质序列综合模型库，完成水质参数从数据空间向序列模型空间的映射；借助于非线性系统估计理论，建立水质序列模型集自适应策略和多模型协同工作机制，完成地下水水质参数的多维预测和水质异常发现；并将污染源反演问题转化为决策变量为污染源位置和强度的最优化问题，建立同步反演处理框架（污染物迁移模型和反演优化模型），改进智能搜索方法，在全局范围内快速搜索问题最优解，实现对地下水突发性污染源的准确定位和快速识别。该课题的研究将为我国地下水污染源的发现、定位及识别提供理论依据和新的解决思路。

项目针对河南省郑州市及周边典型区域，开展地下水水质异常模式发现与分级预警研究。项目以数据分析、机器学习和深度学习方法为工具，以地下水水质监测序列为研究对象，从数据处理的角度出发，关注数据分布特性、提取数据关联关系、分析数据变化趋势、预测数据未来走向，进而完成对地下水环境的认识、分析、鉴别和预警。项目一方面从理论上深入研究符合地下水水质序列特征的序列挖掘、机器学习和深度学习方法，在研究过程中对相关理论进一步深化拓展，为地下水水质异常模式的发现和分级预警提供理论依据。另一方面从技术方法上以水质序列建模->水质序列分析->水质序列预测->水质序列异常发现->分级预警机制构建->水质异常模式分级预警为主线，为地下水水质异常发现和分级预警问题的解决提出了一种新的研究思路。