Abstract: Seismic attributes are often used for structural interpretation and prediction. In order to overcome the problems of multiple solutions and uncertainty caused by single seismic attribute prediction, seismic multi-attribute fusion technology is used to interpret and predict geological structures. Based on the classical machine learning random forest algorithm model, an improved random forest algorithm is proposed to fuse and classify multiple seismic attributes. Combining the seismic multi-attribute fusion technology with the improved random forest algorithm, a geological structure recognition model based on the improved random forest algorithm is established. Taking the second mining area of the second belt of Shanxi Xinyuan Coal Co., Ltd. as the research area, based on the twelve seismic attributes extracted from the three-dimensional seismic exploration results, through the attribute correlation analysis and feature importance analysis of the twelve attributes, according to the results, all twelve attributes are retained for subsequent attribute fusion. Using the exposed and verified geological structure faults and collapse columns as sample labels, an improved grid search optimization algorithm is proposed. The number of classifiers and the maximum feature number of a single decision tree are combined to search the grid. The algorithm model is established based on Python language platform. The experimental results show that the prediction accuracy of the improved algorithm model reaches 97%, After subsequent model verification, it is proved that compared with several algorithms such as logistic regression, gradient lifting and decision tree, the improved random forest algorithm can more effectively identify abnormal bodies such as faults and collapse columns in geological structures, with higher recognition accuracy and wider applicability.