An AVO (Amplitude Variation with Offset) type discrimination method based on a proximal support vector machine provided by the invention includes the following steps: building a parameter model according to the logging data of a target area, and generating a training model parameter table, wherein there are multiple categories for the parameter model, and the training model parameter table includes the category of the parameter model and the attribute parameters of the overlying medium and the attribute parameters of the gas-bearing sandstone layer of the target area under the corresponding category; calculating the basis pursuit inversion reflection coefficient of the target area corresponding to each category of the parameter model, and forming an AVO curve through a fitting function; extracting morphological characteristic parameters from the AVO curve of the target area to form a training set, wherein the morphological characteristic parameters include the monotonicity of the curve, the convexity-concavity of the curve, the number and positions of extremum points and the number and positions of inflection points; and training the training set based on a proximal support vectormachine, and classifying the AVO curve of the target area so as to discriminate the AVO type of the target area. The method not only improves the accuracy of AVO type discrimination, but also reducesthe workload.