| As exploration targets gradually shift from structural reservoirs to lithological reservoirs,the accuracy requirements for reservoir prediction are increasing.Since the 1990 s,more and more machine learning algorithms have been introduced into the prediction of reservoir parameters,such as neural network algorithm and support vector machine algorithm.The neural network algorithm has a large amount of computation and low computational efficiency,and the support vector machine algorithm needs to adjust more parameters.The random forest algorithm is a collection of multiple decision trees,which can well avoid these defects.Compared with the traditional decision tree algorithm,the random forest algorithm has better classification effect,higher prediction accuracy,and is not easy to over-fitting.It is widely used in computer vision,medical and financial aspects,but less in geophysical and reservoir prediction.Based on the classification algorithm of random forest,this paper explores the nonlinear relationship between seismic attributes and reservoir parameter categories.Firstly,this paper studies the basic principles,properties and implementation process of random forest classification algorithms,focusing on the feature selection and the process of obtaining the importance of variables.Secondly,the seismic attributes related to lithology are extracted.Based on the optimal principle of seismic attributes,a subset of sensitive seismic attributes with strong correlation with reservoir parameter categories is selected.The correlation between the obtained seismic attributes is analyzed to ensure the independence of seismic attributes.Subsequently,this paper studies the impact of sample diversity on classification effects and the processing method of unbalanced data sets.The advantages and disadvantages of various resampling methods are discussed,and the smote algorithm is preferred to balance the sample data.Also,evaluation indicators for classification issues are discussed.At the end of the paper,the classification algorithm of random forest is used for actual seismic data,and the reservoir class of the target interval is tested and achieves good results.The research results show that the smote algorithm can effectively eliminate the imbalance of sample data and improve the classification performance of the classifier.At the same time,the random forest classification algorithm can effectively classify the seismic reservoirs and effectively extend the reservoir classification at the well points to the entire target block. |
