The Baiyun Deepwater Zone is located on the front edge of the northern slope of the South China Sea, and multiple gas fields have been discovered, making it a favorable zone for oil and gas accumulation. However, with the deepening of exploration, the seismic exploration targets are becoming increasingly complex, and the uncertainty of fluid properties is one of the current challenges in exploration. Strengthening fluid prediction research is urgent. However, the sensitivity of conventional elastic parameters to gas saturation is weak, and they are influenced by factors such as compaction, lithology, thickness tuning, porosity, etc. The seismic response of gas layers, low saturation gas layers, and good physical property water layers is similar. Conventional elastic parameter inversion methods are difficult to use to solve the problem of gas saturation prediction. Poisson's impedance is manifested in the form of velocity difference between longitudinal and transverse waves, which can eliminate the information of solid phase protrusion and liquid phase, and has fluid factor properties. However, conventional methods of obtaining Poisson's impedance through coordinate rotation have significant errors in quantitatively predicting gas saturation. The ray elastic impedance can be expressed as a generalized Poisson impedance, which can be quickly obtained by stacking the corresponding Poisson angle part of the pre-stack gather data, while retaining frequency characteristics compared to conventional Poisson impedance. Under the guidance of phase controlled rock physics, frequency division technology is adopted to describe the degree of dispersion caused by fluid using frequency dependent Poisson impedance, eliminate the false identification of fluid caused by the use of single amplitude information, and enhance the accuracy and sensitivity of oil and gas detection. At the same time, a nonlinear correlation between characteristic frequency Poisson impedance and gas saturation is established through a random forest algorithm, and porosity controlled volume is added to achieve quantitative oil and gas prediction in effective reservoirs. The random forest algorithm is an integrated algorithm based on decision trees, which has the advantages of fewer adjustment parameters, convenient operation, and good noise resistance. In view of the gas bearing characteristics of the Pearl River Formation gravity flow sand body reservoir in the Baiyun deep-water area, this technology is used to predict the oil and gas bearing properties. The prediction results are highly consistent with the wells, which can effectively distinguish the gas free sandstone and gas bearing sandstone with strong amplitude characteristics affected by physical properties, and verify the effectiveness and application prospects of the method.