Predicting the components and types of kerogen in shale by combining machine learning with NMR spectra

• Combining machine learning with NMR spectra to predict the kerogen components and types. Propose a solution convert 2D graph into available form for models. Build an automatic labeling platform generate over 22,000 training samples. The prediction accuracy of indicators in optimal model can achieve 90% total. method has advantages intelligentization, high-throughput prediction, high accuracy. This study aims develop new that combines nuclear magnetic resonance (NMR) Kerogen is primary hydrocarbon source shale oil/gas, nearly half hydrocarbons are adsorbed kerogen. adsorption generation capacity directly related its types, molecular components, structures. Fruitful researches studying at level have been conducted. Unfortunately, these methods complicated, time-consuming, labor-intensive. Our accuracy, time savings compared existing methods. Additionally, this simplifies operations from repetitive trial error. proposes non-uniform two-dimensional (2D) uniform one-dimensional (1D) matrix, which makes data An constructed annotated groups organic matter molecules their spectra. results show carbon, hydrogen, oxygen element reach 96.1%, 94.8%, 81.7%, respectively. In addition, three types approximately These reflect excellent performance method. Therefore, our work provides automated intelligent analysis method, powerful superior tool studies level.