Production capacity prediction of fractured oil wells based on improved Convolutional Neural Network

Abstract The significance of predicting the productivity of fractured oil wells for optimizing production strategies and enhancing efficiency is widely acknowledged. However, traditional algorithms often fall short in accurately predicting productivity, especially when dealing with fractured oil wells with limited production data, significant data fluctuations, and missing values. An algorithm for predicting fractured oil well productivity based on improved convolutional neural networks(CNN) is proposed in this study. This algorithm not only enables adaptive feature extraction but also enhances prediction performance on small sample data. Experimental results demonstrate that compared to classic CNN and MLP models, the proposed algorithm achieves superior performance. The average absolute errors of initial oil production and incremental oil production are 0.023t and 15.04t, respectively, with corresponding root mean square errors of 0.0011t and 23.74t. The prediction accuracy rates stand at 95.62% and 86.68%, respectively. These performance indicators outperform those of classic models, showcasing promising predictive capabilities. This research holds practical significance in evaluating fractured oil well productivity, devising optimization strategies, and guiding oilfield development endeavors to improve resource utilization efficiency and overall production outcomes.