As a key parameter characterizing the physical properties of reservoir fluids, crude oil density plays an important indicative role in the comprehensive evaluation of oil and gas reservoirs. Traditional laboratory analysis methods suffer from time lag, while existing logging-while-drilling identification technologies excessively rely on rock pyrolysis parameters, which are limited by insufficient parameters and inadequate information dimensions, thus failing to meet the needs of efficient exploration. To address these challenges, this paper takes 115 sets of typical thermal evaporation of gas chromatography samples from Bohai Oilfield as research objects (density range: 0.7542~1.0077 g/cm³), and innovatively constructs a quantitative characterization system for thermal evaporation hydrocarbon gas chromatography. Multidimensional characteristic parameters are extracted through digital processing of the spectra, and the internal coupling relationship between the characteristic parameters and their derived variables and crude oil density is deeply explored via machine learning. The sample dataset was divided into a training set (80%) and a test set (20%) using the stratified random sampling, and a high-precision crude oil density prediction model was flnally established. The results show that the model exhibits excellent prediction performance both on the test set and practical application cases. The mean absolute error between the predicted and measured values is less than 0.02, indicating a high prediction accuracy and reliability. Compared with traditional methods, this technique does not depend on post-completion field sampling and testing, and can quickly realize the quantitative calculation of crude oil density based on cuttings samples during drilling. It provides key technical support for on-site decision-making in oil and gas exploration, with remarkable engineering application value and promation prospects.