中图分类号：TP18；TE341 文献标志码：A DOI: 10.16157/j.issn.0258-7998.256701 中文引用格式： 徐振，张望，李兴亮，等. 基于SAE特征优选和Bagging集成学习的油藏初期产能预测[J]. 电子技术应用，2026，52(3)：84-90. 英文引用格式： Xu Zhen，Zhang Wang，Li Xingliang，et al. Prediction of initial reservoir productivity based on SAE feature optimization and Bagging ensemble learning[J]. Application of Electronic Technique，2026，52(3)：84-90.
Prediction of initial reservoir productivity based on SAE feature optimization and Bagging ensemble learning
Xu Zhen1，Zhang Wang1，Li Xingliang1，Long Jun2
1.Petrochina Tuha Oilfield Branch；2.Shenzhen Pengrui Information Technology Co.， Ltd.
Abstract： The initial productivity of oil reservoir is affected by many factors such as geology, engineering and development, which is a complex nonlinear time series. The traditional prediction method using a single model has low prediction accuracy and weak data adaptability. A prediction model of initial reservoir productivity based on sparse auto encoder (SAE) and ensemble learning is proposed. Firstly, SAE is used to analyze various factors affecting the initial production, and the five-dimensional characteristics of oil saturation, effective thickness of perforated interval, fracturing sand addition, sand addition intensity and energy retention state are automatically extracted as the main control factors. Then, the prediction models are established by using linear regression (LR), support vector regression (SVR) and long short-term memory neural network (LSTM) to predict the initial production. Finally, Bagging ensemble learning is used to comprehensively integrate the prediction results of the three methods, so as to obtain the final prediction results of the initial production of the reservoir. Using the actual data of an ultra-low permeability oilfield to carry out the verification test, the results show that compared with the single LR, SVR and LSTM methods, the proposed method has higher prediction accuracy, stronger data adaptability and higher application prospect.
Key words : sparse autoencoder；feature selection；integrated learning；productivity prediction；support vector regression
