Research on Prediction Model of Fracture Width in Loss Formation Based on Artificial Neural Network

Abstract

Lost circulation in fractured formation has always been a worldwide technical problem in oil and gas drilling projects at home and abroad. The effect of bridging plugging depend on the matching of the fracture width and the particle size of lost circulation material. The plugging formula can be optimized according to the bridge rules to improve the success rate of plugging. In this paper, a prediction model of fracture width based on artificial neural network was established. Input layer, hidden layer and output layer were included. 17 input parameters such as well depth, formation, density of drilling fluid and displacement were determined. The fracture width was the output parameter. By optimizing the number of hidden layers and their nodes, the optimized training set had 2 hidden layers, and the number of nodes in the first hidden layer was 4, and the number of nodes in the second hidden layer was 10, the prediction accuracy of the model was the best, with the RMSE of 1.24 and the maximum R2 of 0.94. The model runs well in the test set with RMSE of 1.28 and R2 of 0.92. The fracture width prediction results of 5 Wells showed that the prediction results of this model match well with the field measurement results, and the average relative error percentage is only 2.94%. The established model had high accuracy, which can provide a basis for early and accurate prediction of well loss risk on site and for guiding optimization of engineering parameters to prevent well loss.