注水开发油田注水通道状态辨识及预测方法

为研究注水开发油田注水通道的状态,通过将油田动、静态数据分析与机器学习算法相结合,提出了一种基于油田勘探开发动、静态数据进行注水井地层吸水状态预测的新方法.首先利用高斯混合模型完成地层吸水状态分类;然后结合动、静态数据生成机器学习样本,采用随机森林算法构建地层吸水状态预测模型,并对影响地层吸水状态的地质因素和开发因素进...

Identification and predictions of water injectivity for water injection channels in water injection development oilfield

This paper presents a new method for predicting water injectivity of the well formation based on dynamic and static data of oilfield exploration and development. Initially, the water injectivity was classified using Gaussian Mixture Model (GMM). On this basis, machine learning samples were generated using dynamic and static data, and further a prediction model of water injectivity was established using the random forest algorithm. Moreover, the importance of geological factors and development factors affecting water injectivity was also analyzed. Finally, the model has been applied to the study area, providing the water injectivity of all wells in different sections and different periods. The predicted results are consistent with the tracer monitoring results and the water absorption profile test results, thus verifying the feasibility of the method proposed in this study. The method overcomes the disadvantages of traditional methods, such as the hypotheses of ideal conditions, experience and parameters, and explores the water injectivity by historical data mining. It has an important guiding significance for profile control and water plugging in water injection development oilfield.