In order to complete the study of seismic fine description and thickness prediction of facies change reservoir, Subei Basin was taken as an example. In the study of the beach bar sand reservoir in Sanhebei area of Subei Basin, with increasing depth, the seismic events changes from the outer slope's "single eveut" to the middle slope's "double strong eveuts" and then to the inner slope's "upper strong and lower weak" reflection. In this study area, there have been 20 wells drilled, and the prediction of the reservoir belongs to a small-sample case. It is essential to understand the mechanism of the facies separation and carry out the description and prediction of the storage layer in each facies separation, which has both theoretical research significance and practical necessity. For the facies zone division, first of all, it is necessary to clarify the essence of the facies change through fine analysis of the relationship between wells and seismic data. First, there are structural changes in the thickness of the overlying layer and the storage layer. Second, there are changes in the composition distribution of gray matter. These two factors combined cause the facies change of the event. Subsequently, by extracting the number of zero crossing points, energy-half-time, maximum amplitude, dominant frequency, bandwidth, and minimum amplitude from the seismic data, the distribution range of the facies change was clarified, and the attribute slices support the division of the study area into three facies separation. Considering that the objective layer is close to the upper covering layer and the sand body features are not distinct in the conventional zero-phase data, compressed sensing frequency enhancement and 90° phase shift target processing were conducted to further clarify the characteristics of the three facies change zones. Regarding the thickness prediction of the facies change reservoirs, first, the thickness is calculated using the attribute and thickness fitting method, but the results were not ideal. Then, SVM(support vector machine) machine learning method was used to predict the thickness without facies separation, and the results indicated that the SVM method was somewhat improved compared to the previous fitting method. Finally, the two methods were used to calculate and predict the thickness of the facies change reservoirs separately, and the accuracy was further improved. The results have some reference value for the exploration and development with few wells and complex facies change reservoir.