Abstract: During the deposition period of ES3 and ES4 in Dongying sag, a large number of glutenite bodies developed in different periods, which are important reservoir types in unconventional oil and gas exploration. Due to the characteristics of large variation in vertical thickness of glutenite body, uneven distribution in the lateral direction, and rapid change of lithofacies, the use of a single seismic attribute in the attribute analysis has great uncertainty in the description of reservoir thickness. To this end, a variety of seismic attributes are extracted, and the principal component analysis method is used to optimize and remove redundant information. Considering that the random forest (RF) has the characteristics of high prediction accuracy, strong tolerance to outliers, fast training speed and not easy to over fit, etc., this algorithm is introduced in this paper to predict the thickness of glutenite reservoirs. For the self-similarity of attributes, PCA adopts two methods:One is to directly reduce the dimension of all attributes and extract the principal components for prediction (PCA-PF1); the other is to first reduce the dimension of similar attributes and then combine other attributes to make prediction(PCA-RF2). The original RF, PCA-RF1, PCA-RF2 are also compared with the artificial neural network (ANN). The results show that the PCA-RF2 method with similar attribute dimensionality reduction has the best application effect.