Reservoir Permeability Prediction Method Based on Fuzzy Clustering and Machine Learning

With the in-depth development of oil and gas exploration technology and the development of low permeability reservoirs, the permeability prediction method represented by machine learning technology has been widely applied. However, it is limited by data distribution, data heterogeneity, and model complexity. Traditional machine learning methods are not accurate, and the results are not stable when predicting permeability. Aiming at some problems encountered by traditional machine learning algorithms in predicting permeability, this paper takes carbonate rocks in Region X as the research object, comprehensively analyzes the geological characteristics of carbonate rocks, analyzes the importance of input features by XGBoost built-in module, and selects seven logging curves as the inputs of the network model combined with actual conditions. A Fuzzy C-means algorithm based on fuzzy logic is used to cluster the data. According to the clustering results, the data is divided into several subsets, and the SVR or LSTM model is trained for each subset. The predictions of all subsets are then combined to get the final prediction. The performance of the model is evaluated using evaluation indicators (such as mean square error, determination coefficient) and compared with direct prediction methods without clustering. The results show that the coefficient of determination (R²) of prediction after FCM clustering reaches 0.86, and the average absolute error is 0.068. The determination coefficients (R²) of LSTM and SVR algorithms for direct prediction are 0.75 and 0.73, respectively. It is proved that fuzzy logic combined with machine learning technology can better characterize the nonlinear mapping relationship between data, and explore more potential relationships between parameters to improve the prediction effect.