基于多元回归KNN的油田缺失数据填充方法

Filling method of missing data in oilfield based on multiple regression KNN

为了提高油田开采的安全性和科学性,油田中装有各型数据传感器,但数据缺失导致传感器采集数据可用性显著降低。针对油田传感器大比例数据缺失填充问题,提出了一种基于多元回归KNN的缺失数据填充方法。该方法首先基于KNN利用传感器数据空间相关性预测缺失值,其次基于多元回归利用传感器数据时间相关性预测缺失值,最后将时空相关性预测结果通过样本决定系数进行整合。分别采用标准数据集和油田传感器数据集进行性能对比实验,结果验证了该方法对缺失数据填充的有效性和准确性。

In order to improve the safety and scientificity of oilfield extraction,various types of data sensors are installed in the oilfield,but the lack of data leads to a significant reduction in the availability of data collected by the sensors.Aiming at the problem of missing filling of large-scale data of oil field sensors,a missing data filling method based on multiple regression KNN was proposed.Firstly,the spatial correlation of sensor data is used to predict missing values based on KNN.Then,multiple regression is used to predict missing values using the temporal correlation of sensor data.Finally,the spatio-temporal correlation prediction results are integrated by sample determination coefficients.Performance comparison experiments are performed by using standard data sets and oil field sensor data sets,respectively.The results verify the effectiveness and accuracy of the method for missing data filling.