Machine learning approach for identification of release sources in advection-diffusion systems

摘要

Mathematically, the contaminant transport in an aquifer is described by anadvection-diffusion equation and the identification of the contaminationsources relies on solving a complex ill posed inverse model against theavailable observed data. The contaminant migration is usually monitored byspatially discrete detectors (e.g. monitoring wells) providing temporal recordsrepresenting sampling events. These records are then used to estimateproperties of the contaminant sources, e.g., locations, release strengths andmodel parameters representing contaminant migration (e.g., velocity,dispersivity, etc.). These estimates are essential for a reliable assessment ofthe contamination hazards and risks. If there are more than one contaminantsources (with different locations and strengths), the observed recordsrepresent contaminant mixtures; typically, the number of sources is unknown.The mixing ratios of the different contaminant sources at the detectors arealso unknown; this further hinders the reliability and complexity of theinverse-model analyses. To circumvent some of these challenges, we havedeveloped a novel hybrid source identification method coupling machine learningand inverse analysis methods, and called Green-NMFk. Our method is capable ofidentifying the unknown number, locations, and properties of a set ofcontaminant sources from measured contaminant-source mixtures with unknownmixing ratios, without any additional information. It also estimates thecontaminant transport properties, such as velocity and dispersivity.