Application of artificial neural network in evaluating the effectiveness of mechanical water shutoff in horizontal wells

Horizontal wells often have to face water breakthrough and correspondingly, a sharp decrease in oil production during water flooding processes. Mechanical water shutoff, recognized as a cost-efficient technique for water control, is complicated by intricate well selection factors. In this work, a novel BP-ANN model was developed to predict the oil production increment of horizontal well after a mechanical water shutoff operation. It aims to select a suitable candidate well for such treatment and reduce the economic risk. Specifically, a multivariate analysis with three different methods is implemented to quantify the relationships between the 15 factors and mechanical water shutoff operations effects and determine the possible dominated factors. Then, a BP-ANN model, consisting of 13 input parameters and four outputs, is developed according to 80 horizontal wells data with mechanical water shutoff. The optimized results show that the reasonable structure of such BP-ANN model should include 21 hidden layer nodes, relu-type activation function, three hidden layers, learning rate of 0.005 and Nadam optimizer. Twenty-three horizontal wells were selected for validation, and the model’s output learning accuracy exceeded 99.0%. The model is verified to be able to provide significant basis for assessing the effectiveness of mechanical water shutoff in horizontal wells.