Intelligent geochemical interpretation of mass chromatograms: Based on convolution neural network

• DOI: 10.1016/j.petsci.2023.11.010
• Publication Date: 2023-11-11T16:54:40Z
• Authors: Kai-Ming Su; Jun-Gang Lu; Jian Yu; Zi-Xing Lu; Shi-Jia Chen
• Keywords: Convolution (computer science)

Gas chromatography-mass spectrometry (GC-MS) is an extremely important analytical technique that widely used in organic geochemistry. It the only approach to capture biomarker features of matter and provides key evidence for oil-source correlation thermal maturity determination. However, conventional way processing interpreting mass chromatogram both time-consuming labor-intensive, which increases research cost restrains extensive applications this method. To overcome limitation, a model developed based on convolution neural network (CNN) link samples from Triassic Yanchang Formation, Ordos Basin, China. In way, can be automatically interpreted. This first performs dimensionality reduction 15 parameters via factor analysis then quantifies using two indexes (i.e. MI PMI) represent parent material type, respectively. Subsequently, training, interpretation, validation are performed multiple times different CNN models optimize structure hyper-parameter setting, with as input obtained PMI values supervision (label). The optimized presents high accuracy chromatogram, R2 typically above 0.85 0.80 interpretation results, significance twofold: (i) developing efficient geochemical research; (ii) more importantly, demonstrating potential artificial intelligence geochemistry providing vital references future related studies.