The invention discloses a method for predicting the underground river filling degree based on GR frequency-divided inversion. The method comprises the steps of performing frequency division
processing on original seismic data by using a Marr
wavelet frequency division technology, acquiring frequency-divided data volumes with different frequency bands, respectively extracting frequency division attributes of different frequency bands from the frequency-divided data volumes, building a kernel function according to an amplitude-frequency relation under different thicknesses of a reservoir, performing multiple times of learning by using a
support vector machine, establishing a nonlinear mapping relation between the frequency division attributes and
well logging GR curves, combining the nonlinear mapping relations between the frequency division attributes of different frequency bands and the
well logging GR curves together to acquire a GR frequency-divided inversion body, determining the GR peak distribution probability corresponding to underground river samples with different filling degrees according to a
well logging interpretation result, and determining the underground river filling degree of the GR frequency-divided inversion body. The scheme can acquire an inversion result with
high resolution, can directly predict muddy filling conditions of a reservoir in an undrilled region, can improve the construction and
production rate of a development and adjustment well and provides
technical support for formulating an oilfield development scheme.