Accurately predicting the underground fracture width is crucial for selecting suitable plugging techniques and materials. However, existing analytical models based on physical mechanisms for fracture width have significant limitations. In this study, multiple underground fracture width prediction models were constructed using convolutional neural network (CNN), long short-term memory (LSTM) neural network, and CNN-LSTM fusion neural network algorithms, combined with geomechanical data. The results showed that the CNN-LSTM model had the best predictive performance on the test set, the correlation coefficient R2 is 0.967, and the Mean Squared Error (MSE) is 0.0005. Moreover, it successfully predicted 150 sample points from well B-2 with an accuracy rate of over 90%. By utilizing the CNN-LSTM model to predict fracture width and optimize the plugging agent formulation, the success rate of on-site plugging in well B-5 in the Bohai oilfield was significantly improved. This application demonstrates that fracture width prediction based on intelligent models can provide reliable decision support for engineers, ensuring the effectiveness of plugging techniques and enhancing the success rate of initial plugging on-site.