Crack Width Prediction Model Based On Optimized BP Neural Network

The problem of fractured well loss is a serious threat to the safety and progress of drilling production. At present, the success rate of field plugging is relatively low one of the important reasons is that the fracture width cannot be accurately predicted, and the limitations of fracture width cognition make it difficult to determine the plugging method and material particle size. Therefore this paper puts forward to use the improved BP neural network method to establish the prediction model of lost circulation fracture width in order to solve the problem of fracture width prediction. Firstly, according to the analysis of variance method, the related parameters that affect the crack width are determined which are input into the improved BP neural network model for training, and the accuracy is verified by the sample data prediction. Finally, the slope of the fitting straight line of the test set data is 0.8772, and the intercept is 0.0206. In addition, in order to confirm the stability of the model, the performance of the crack width prediction model is evaluated, and the model's determination coefficient (R²) is 0.89, the average absolute percentage error (PCC) is 0.82, and the root mean square error (RMSE) is 1.35, which proves that the performance of the model is excellent. Finally, the field engineering data is used for example prediction. From the results, it can be seen that the model has high prediction accuracy and can provide better assistant decision-making in plugging engineering operation.